Most lists of the best MCP servers for finance data stop at what each one does. The questions that decide a purchase come later: which plan unlocks the MCP server, how long a free key lasts when an agent makes eight calls to answer one question, and whether the dataset you need sits two tiers up.

This ranking lines up five provider-run servers on exactly those points. Each server gets one section with the same four parts (data coverage, pricing, setup and limits), so you can compare them line by line. Prices are personal-use list prices in US dollars from each provider’s published plans in October 2026.

How these finance MCP servers were compared

Every server here is run by its own data provider. Community wrappers were left out because their pricing and limits depend on someone else’s API.

Each one was checked on five things:

• Data coverage: which datasets the server can reach, and on which plan
• Cost: the free option, the cheapest plan that includes MCP access, and the plan that unlocks the full dataset
• Setup: hosted or local, how sign-in works, and whether Claude lists it as a ready connector
• Limits: request caps, credits, bandwidth and data-source caveats an agent will hit in practice
• Output type: raw data fields you can check, or a written answer from the provider’s own agent

The last point matters more than it looks. Four of these servers return numbers your agent works with directly. Octagon returns an answer its own agent has already written, which changes both what you pay for and how you verify it.

Finance MCP servers at a glance

1. Alpha Vantage: best overall, with raw data from one tool per API function and the broadest market coverage.
2. FMP: best overall for company data, with tools grouped by dataset for statements, filings and transcripts.
3. Octagon: written answers from research agents, strongest for due diligence and Kalshi markets.
4. FinancialData.Net: raw data from one tool per documented endpoint, strongest for traceable fundamentals and ownership.
5. EODHD: raw data from 72 tools, strongest for non-US exchanges and macro series.

1. Alpha Vantage: best overall, with the widest coverage on a free key

Alpha Vantage and FMP share the best-overall spot. Alpha Vantage takes first place because one official connector reaches almost every dataset in this comparison, and the free key works with it.

Data coverage. US stocks, US options with Greeks, financial statements, earnings call transcripts with sentiment, insider and institutional holdings, economic indicators, forex and crypto. By Alpha Vantage’s count, that is more than 130 data functions.

Pricing. The free key allows 25 requests a day. Premium runs from $49.99 a month for 75 requests a minute to $249.99 for 1,200, and paid plans have no daily cap.

Setup. It is listed in Claude’s Connectors directory, so it connects without a URL. The hosted endpoint is mcp.alphavantage.co/mcp with OAuth sign-in, and uvx marketdata-mcp-server runs it locally.

Limits. The free key runs out fast once an agent starts chaining calls. Realtime and 15-minute delayed US prices need the right premium entitlement, so check which feed your key returns before relying on intraday numbers.

2. Financial Modeling Prep (FMP): best overall for company data, priced by plan tier

FMP is the other best-overall pick. No other server here goes as far into statements, filings and valuation, though the plan decides which of that data you actually get.

Data coverage. Statements and ratios, 10-K and 10-Q data, DCF valuations, analyst estimates, insider trades, Treasury rates and economic indicators, plus prices, forex and crypto. Premium adds UK and Canada, and Ultimate adds global coverage, earnings call transcripts and 13F holdings.

Pricing. The free Basic plan gives 250 calls a day and works with MCP. Starter is $29 a month ($19 billed yearly), Premium $69 ($49 billed yearly) and Ultimate $139 ($99 billed yearly), at 300, 750 and 3,000 calls a minute.

Setup. Hosted at financialmodelingprep.com/mcp with the API key in the URL. Add it in Claude as a custom connector; there is nothing to install.

Limits. Starter is US-only, with five years of history and annual fundamentals. Every plan also has a trailing 30-day bandwidth cap, from 500MB on the free plan to 150GB on Ultimate, and bulk pulls can hit it before the call limit.

3. Octagon: research answers, paid by the credit

Octagon is third because it sells something different: answers its own agents have already written, not raw fields.

Data coverage. Agents for SEC filings (10-K, 10-Q, 8-K, S-1, 13F and more), earnings call transcripts, stock and crypto data, ownership and news, plus a deep research agent. It is also the only server here with Kalshi prediction markets.

Pricing. There is no free plan, only a 15-day trial. Plus is $17 a month billed yearly for 200 credits, and Pro is $41 billed yearly for 600. A public-market call costs 1 credit, deep research 2 and a prediction-market call 3, and extra credits are $50 per 500.

Setup. Hosted at mcp.octagonagents.com/mcp with OAuth, added as a custom connector. npx -y octagon-mcp runs it locally.

Limits. Treat any number in an answer as a lead and check it against the filing before reusing it. Prediction-market questions use credits three times as fast as company questions.

4. FinancialData.Net: widest dataset, MCP from $149 a month

On data alone, FinancialData.Net would sit higher. It is fourth because MCP access starts at $149 a month, with no free way to try it first.

Data coverage. US and international stocks, option chains with Greeks, futures, statements and ratios including international filers, insider and institutional trading, Senate and House trades, ETF and mutual fund holdings, ESG scores and economic data.

Pricing. The free tier gives 300 REST requests a day without MCP. MCP starts on Professional at $149 a month or $1,299 a year, at 30 requests a second. Enterprise is $299 a month and adds external commercial use and redistribution.

Setup. Hosted at financialdata.net/mcp with the key in the URL, added as a custom connector. An open-source local server runs with uv.

Limits. You can only test the MCP server after paying for Professional. Use the free REST tier to check the data first.

5. EODHD: cheapest route to international stock data

EODHD gives the most non-US coverage for the money. A caveat on its price data keeps it fifth for trading use.

Data coverage. End-of-day, intraday and live prices across exchanges using SYMBOL.EXCHANGE tickers, US tick data, fundamentals with a bulk tool for up to 500 tickers per call, Form 4 insider filings, news sentiment, Treasury yields, macro data and an economic events calendar. That adds up to 72 read-only tools.

Pricing. A free key works with limited data. Data packages start at $19.99 a month each, and the All-In-One package is $99.99 a month or $999.90 a year.

Setup. For Claude, use the OAuth endpoint at mcp.eodhd.com/v2/mcp. A v1 endpoint takes an API key instead, and both servers are open source.

Limits. EODHD calls its prices indicative, aggregated with VWAP across many sources rather than taken straight from exchanges. That works for screening and fundamentals and less well for intraday execution. US options are a separate marketplace add-on.

Questions about cost and choice

How much does an MCP server for financial data cost?

Four of the five have a free plan or a trial. Paid entry points run from Octagon’s Plus plan at $17 a month billed yearly up to FinancialData.Net’s Professional plan at $149 a month, the cheapest plan that includes its MCP server.

Do finance MCP servers cost extra on top of the data plan?

For Alpha Vantage, FMP and EODHD, no: the MCP server uses the same key and the same plan limits as the API. FinancialData.Net only includes MCP from its Professional plan upward, and Octagon charges credits for every agent call.

Which MCP server is cheapest for international stock data?

EODHD. Its free key works with limited data and paid packages start at $19.99 a month. FinancialData.Net includes international stocks but needs the $149 Professional plan for MCP, and FMP’s global coverage comes with Ultimate.

MCP server pricing: what an agent actually spends

One MCP tool call is one API request. Assume a question such as “compare Apple and Microsoft margins over five years” takes eight calls; the real number depends on the agent and the prompt. At that rate, Alpha Vantage’s free key covers about three questions a day (25 ÷ 8), and FMP’s free plan about 31 (250 ÷ 8).

Octagon’s credits stretch differently. Plus’s 200 monthly credits cover 200 company questions, or about 66 prediction-market questions.

All prices above are personal-use list prices in USD as of October 2026, from the pricing pages of Alpha Vantage, FMP, Octagon, FinancialData.Net and EODHD. Commercial licences cost more.

Which finance MCP server to pay for, by budget

• $0: FMP’s free plan for company data at 250 calls a day, Alpha Vantage’s free key for broader market data at 25 a day, or EODHD’s free key to test non-US tickers.
• Under $50 a month: Octagon Plus at $17 billed yearly, FMP Premium at $49 billed yearly, Alpha Vantage’s first premium tier at $49.99, or single EODHD packages from $19.99.
• $99 to $149 a month: FMP Ultimate at $99 billed yearly, EODHD All-In-One at $99.99, or FinancialData.Net Professional at $149.

Best MCP Servers for Finance Data in 2026 was originally published in DataDrivenInvestor on Medium, where people are continuing the conversation by highlighting and responding to this story.

Computing the ratios is the easy part of selling a Shariah stock screener. Any spreadsheet can divide debt by market capitalization. The hard part is explaining a verdict when two apps disagree about the same stock, and they often do. AAOIFI measures debt against market capitalization, MSCI measures it against total assets, and every provider decides for itself what counts as debt, cash and impure income.

This article builds the core of a screening engine on Financial Modeling Prep data, using Claude through the FMP MCP server. It screens 12 US stocks as of the October 16, 2026 close in three steps: a business activity screen, a financial ratio screen under AAOIFI Shari’ah Standard №21 with an MSCI comparison, and a monitor that shows how October’s price drop moved debt ratios. Every verdict carries the ratio, the threshold, the data date and a reason, so a reviewer can trace how it was reached.

The output is a screening record, not a fatwa. The engine does not decide whether a business line is permissible, which assets count as Shariah-compliant, or how lease obligations should be treated. Those calls belong to a Shariah supervisory board, and the article marks each place one is needed.

Key takeaways

• Of the 12 stocks, JPMorgan Chase and Las Vegas Sands were excluded at the business activity screen, and Costco and Tesla were sent to business review.
• Under AAOIFI, 5 of the remaining 10 passed all three ratio tests. Oracle (41.00%) and Charter Communications (731.2%) failed on debt to market capitalization.
• On MSCI’s total-assets basis, only Microsoft, Amazon and Meta passed outright. Apple, NVIDIA and Alphabet moved to review once long-term investments were counted as interest-bearing.
• FMP returned zero interest income for Apple and Charter because neither reports it as a separate line. The engine records that test as Not reported instead of passing it.
• Oracle’s debt ratio rose from 38.29% to 41.00% between September 30 and October 16 with no change in reported debt, which is why a screener needs a price-driven re-screen.

Building the Shariah screening framework

The 12 stocks were chosen by the author to send at least one company down every path the engine has: large-cap technology names, a heavily indebted cable operator, a bank, a casino operator and a retailer that sells alcohol and tobacco. Every value comes from FMP as of the October 16, 2026 close, in US dollars.

Standards used

AAOIFI is the primary rule set. MSCI’s Islamic Index Series is the comparison, because it uses total assets as the denominator instead of market capitalization.

The AAOIFI thresholds follow the summary in this 2025 study of Shariah screening standards. MSCI also tests receivables plus cash against total assets; that ratio was left out because MSCI has consulted on raising its threshold, and a product should read the current methodology before adding it.

Business activity rules

Industry labels and keywords are a first filter. They can’t measure what share of revenue a prohibited activity contributes, so a keyword hit sends the company to review instead of excluding it.

Ratio measures

Decisions a Shariah board must approve

• Leases: FMP’s totalDebt includes lease obligations. Microsoft's is $128.81 billion against $40.29 billion of borrowings. The engine tests borrowings only.
• Cash: FMP doesn’t split interest-bearing from non-interest-bearing balances, so all cash and short-term investments are treated as interest-bearing. That is the conservative reading.
• Impure income: interest income is the only non-permissible income FMP reports as a line item. Revenue from prohibited business lines needs segment or filing review.
• Market cap: AAOIFI is tested against a single closing value. S&P and the MSCI M-Series use multi-year averages, which this engine doesn’t compute.

Status and missing-data rules

These are the author’s screening rules for routing companies, separate from any accounting materiality standard.

If a call fails, Claude retries once. A missing value is never replaced with an estimate or a value from another source.

FMP data inputs and access through Claude MCP

All inputs come from four FMP operations. US fundamentals in FMP are sourced from SEC EDGAR filings, so each balance sheet row carries the period end and filing date the engine stores with every verdict.

The industry labels in Table 1 match FMP’s own list from the directory → available-industries operation, so the exclusion rule can't miss a label because of a spelling difference.

Scope and units. Values are consolidated, in US dollars, shown in millions unless marked otherwise. Ratios are percentages to two decimals. The balance sheet is the latest quarter filed on or before October 16, 2026, and the TTM row is the one whose date matches it. Market cap is the October 16, 2026 close from the historical series, so a later run returns the same number. The live profile value changes every trading day.

Connecting FMP to Claude

1. Get an API key from your FMP account.
2. In Claude, open Settings → Connectors and add a custom connector with the URL from the FMP MCP server documentation, including your key.
3. Start a new chat, enable the FMP connector and run the prompts below in order.

Retrieval sequence

1. Profile for each of the 12 symbols, then the business activity rules.
2. For every company not excluded: quarterly balance sheet, TTM income statement and the market cap series.
3. Ratios and verdicts from the October 16 values.
4. The September 30 values from the same market cap series for the re-screen monitor.

Running the screening engine in Claude

The work is split into three prompts because each one maps to a module a product would run on its own schedule. The business screen changes rarely and decides which companies need financial data at all, so it runs first and saves calls. The ratio screen reruns when a new quarterly filing arrives. The monitor reruns on price moves, which can push a company across a threshold between filings.

All prompts were run on October 17, 2026, against data as of the October 16, 2026 close. Amounts are in US dollar millions.

Prompt 1: Business activity screen

The first prompt fixes the universe and applies the industry and keyword rules from the framework. It returns one table, so a reviewer sees each company’s label next to the rule that decided it.

Objective: Apply a Shariah business activity screen to 12 US stocks using FMP company profiles.

Universe: AAPL, MSFT, NVDA, GOOGL, AMZN, META, TSLA, ORCL, COST, CHTR, JPM, LVS.

Data rules:
- Call company → profile-symbol once per symbol. Retry a failed call once. Do not use any other source.
- Use the industry and description fields exactly as returned.

Steps:
1. Mark Excluded if the industry is one of: Banks; Banks - Diversified; Banks - Regional; Financial - Credit Services; Financial - Mortgages; Financial - Capital Markets; Investment - Banking & Investment Services; any industry starting with "Insurance"; REIT - Mortgage; Beverages - Alcoholic; Beverages - Wineries & Distilleries; Tobacco; Gambling, Resorts & Casinos.
2. If not excluded, search the description for: alcoholic, tobacco, casino, gambling, pork, adult entertainment, insurance. Any match = Review Required.
3. Otherwise mark Pass.

Output: Table 1. Business Activity Screen, with columns Symbol | Company | Sector | Industry | Result | Reason. Keep the order of the universe list.

Limits: Do not judge revenue share or permissibility. A keyword match only routes the company to review.

What Prompt 1 returned

Claude made 12 profile calls with no failures. Two companies were excluded and two were routed to review.

Table 1. Business Activity Screen

Costco’s industry label, Discount Stores, would have passed an industry-only screen. The description check caught the alcohol and tobacco sales, which is the case a product has to handle when a retailer’s prohibited revenue is a small share of a large total. Tesla’s flag is weaker, since insurance is a minor service line, but the rule routes both to the same review queue.

Prompt 2: Financial ratio screen

The second prompt runs the ratio tests on the 10 companies Prompt 1 didn’t exclude, including the two in business review, so a reviewer has the full financial picture. It separates inputs from verdicts so every ratio can be checked against the values beneath it.

Objective: Screen 10 stocks against AAOIFI Shari'ah Standard No. 21 and the MSCI Islamic Index Series total-assets ratios using FMP data.

Universe: AAPL, MSFT, NVDA, GOOGL, AMZN, META, TSLA, ORCL, COST, CHTR.

Data rules:
- Balance sheet: statements → balance-sheet-statement, period=quarter, limit=2. Use the latest row with filingDate on or before 2026-10-16.
- Income: statements → income-statements-ttm, limit=2. Use the row whose date equals the balance sheet date.
- Market cap: company → historical-market-cap, from_date=2026-09-30, to_date=2026-10-16. Use the 2026-10-16 value.
- Retry a failed call once. Never estimate or substitute a value.

Definitions:
- Borrowings = shortTermDebt + longTermDebt. Report lease obligations (capitalLeaseObligations) separately; do not include them.
- Cash = cashAndShortTermInvestments.
- Impure income share = interestIncome ÷ revenue. If interestIncome = 0 and cash > 0, mark the test Not reported.

Steps:
1. AAOIFI: Borrowings ÷ market cap < 30%; Cash ÷ market cap < 30%; Impure income share < 5%.
2. MSCI: Borrowings ÷ totalAssets < 33.33%; Cash ÷ totalAssets < 33.33%. Also compute (Cash + longTermInvestments) ÷ totalAssets.
3. Verdict per standard: Fail if any test is at or above its threshold. Review Required if no test fails but a test is Not reported, the company is in business review, or (MSCI only) Cash + longTermInvestments ≥ 33.33%. Otherwise Pass.

Output:
- Table 2. Screening Inputs: Symbol | Balance sheet date | Borrowings | Lease obligations | Cash and ST investments | Long-term investments | Total assets | TTM revenue | TTM interest income | Market cap (Oct 16, 2026). USD millions.
- Table 3. AAOIFI and MSCI Verdicts: Symbol | Debt / market cap | Cash / market cap | Interest income / revenue | AAOIFI | Debt / total assets | Cash / total assets | Cash + LT investments / total assets | MSCI | Reason. Percentages to two decimals.

Limits: Do not classify individual assets or debts as Shariah-compliant. Do not test income from prohibited business lines.

What Prompt 2 returned

Claude made 30 calls, three per company, with no failures. Every balance sheet was filed before the cut-off date, and every TTM row matched its balance sheet date.

Table 2 holds the raw inputs. Apple and Charter show zero interest income, and the lease column shows how much FMP’s totalDebt would have added.

Table 2. Screening Inputs (USD millions)

Table 3 applies the thresholds. The two standards agree on the failures but split on NVIDIA and Alphabet.

Table 3. AAOIFI and MSCI Verdicts

Under AAOIFI, five companies passed, three need review and two failed. Under MSCI, only three passed. The difference comes from the denominator: NVIDIA’s cash is under 2% of its market cap but 31.03% of its total assets, and 47.00% once long-term investments are added. Whether those investments are interest-bearing is a question the balance sheet line can’t answer, so the engine flags it instead of guessing.

Prompt 3: Price-driven re-screen monitor

AAOIFI’s debt test moves with the share price even when the balance sheet doesn’t. The third prompt compares the debt ratio at two closes and measures how far market cap would have to move to cross 30%, which tells a product which holdings to re-screen first.

Objective: Show how market cap changes moved each company's AAOIFI debt ratio between two dates.

Universe: The 10 companies from Prompt 2.

Data rules:
- Use borrowings from Table 2. Do not refetch the balance sheet.
- Use the 2026-09-30 and 2026-10-16 values from the historical-market-cap series already retrieved.

Steps:
1. Debt ratio at each date = borrowings ÷ market cap.
2. Change = Oct 16 ratio minus Sep 30 ratio, in percentage points.
3. Market cap move to reach 30% = (borrowings ÷ 0.30) ÷ Oct 16 market cap − 1.
4. Priority: Breach if the Oct 16 ratio is 30% or more; Watch if it is 25% or more, or rose by 1 point or more; otherwise Routine.

Output: Table 4. Debt Ratio Re-Screen Monitor: Symbol | Market cap (Sep 30) | Market cap (Oct 16) | Debt / market cap (Sep 30) | Debt / market cap (Oct 16) | Change (pts) | Market cap move to reach 30% | Priority. USD millions; percentages to two decimals, move to one decimal.

Limits: This table does not change the Prompt 2 verdicts. It only ranks companies for re-screening.

What Prompt 3 returned

No new calls were needed. Oracle and Charter are already in breach, and every other company would need a market cap fall of more than 80% to reach the line.

Table 4. Debt Ratio Re-Screen Monitor

Oracle shows the effect the monitor exists for. Its borrowings stayed at $155.93 billion, yet a 6.6% fall in market cap added 2.71 points to its debt ratio. To pass again on the same balance sheet, its market cap would need to rise 36.7%. For the large technology names, the debt test is far from the line, so their re-screen can wait for the next filing.

Reading the screening results

The clearest outcomes are the ones a customer would expect: a bank and a casino operator excluded on business activity, and two heavily indebted companies failing the debt test under both standards. A product can show those verdicts with little explanation.

The review queue is where a screener earns its price. Of the 10 companies tested, AAOIFI left three in review and MSCI left five. None of those verdicts can be settled from the structured data:

• Costco and Tesla need revenue figures for alcohol, tobacco and insurance. The next step is the segment note in each company’s 10-K, measured against the 5% limit.
• Apple and Charter need interest income from the notes to the financial statements, since neither shows it as a separate line.
• NVIDIA, Alphabet and Apple need their long-term investments split into interest-bearing securities and equity stakes before the MSCI cash test can be closed.

Two results also need a data check before they go live. Charter’s October 16 market cap of $13.23 billion gives a debt ratio above 700%, and a value that extreme should be confirmed against shares outstanding in the latest filing. Tesla’s profile showed a market cap of $1.40 trillion while the historical series gave $1.23 trillion for October 16; the engine uses the dated series, but a product displaying both would confuse users.

When analyst review is required

• A business keyword matches, whatever the ratios show
• Interest income is Not reported while the company holds cash or investments
• MSCI cash passes on short-term holdings but fails once long-term investments are added
• A debt ratio moves into the Watch band between filings
• Two FMP endpoints return different values for the same field and date

Turning the screening record into a product

The three prompts are the engine’s modules. What turns them into something a customer pays for is the record behind each verdict. For every company and standard, store the ratio values, thresholds, input values, balance sheet date, filing date, market cap date, rule version and the reason text from Table 3. When a customer asks why your app disagrees with another one, that record is the answer.

A few things sit outside the engine and need settling before launch:

• Shariah approval. A supervisory board should sign off on the rule set, the lease and cash treatment, and the review workflow, and the app should name the standard behind each verdict.
• Refresh schedule. Rerun Prompt 2 when a new quarterly filing lands, and Prompt 3 daily for any company in the Watch or Breach band.
• Data rights. A paid screener displays FMP-sourced values to its users, which needs more than a personal plan. FMP’s guide to using its data in a public app or client dashboard separates display, redistribution and derived outputs, and its terms of service define commercial use. Settle the licensing before the first customer sees a verdict.

The next build step is the review queue: pulling segment revenue and investment notes from filings for the companies in Table 3 marked Review Required, so the engine can close more verdicts without a person reading every 10-K.

Build a Shariah-Compliant Stock Screening Engine With FMP MCP was originally published in Level Up Coding on Medium, where people are continuing the conversation by highlighting and responding to this story.

Two copy-ready prompts that check six holdings against AAOIFI’s debt, cash and interest tests and work out how much of a year’s dividends to purify.

When a stock you own drops 10% in a week, most investors check the price. A Muslim investor has a second question: is it still Shariah-compliant? The answer can change without the company doing anything. Under the AAOIFI standard, a company’s debt is compared with its market value, so a falling share price makes the same debt look bigger.

This article shows how to check your own holdings with Claude and FMP data, using two prompts you can copy. The sample portfolio holds six US stocks: Microsoft, NVIDIA, Meta, Johnson & Johnson, Procter & Gamble and Oracle. Claude checks each one against AAOIFI’s three financial tests on the September 30 and October 16, 2026 closes, then works out how much of the past year’s dividends to give away as purification, using the two methods investors most often follow.

You don’t need to code or read a balance sheet. You do need to know which standard your scholar or fund follows, because the results only answer AAOIFI’s tests. Whether a business line is permissible, and which purification method applies to you, are questions for a scholar.

Nothing here is investment or religious advice.

Key takeaways

• Five of the six holdings passed all three AAOIFI tests on October 16, 2026. Oracle failed, with debt equal to 41.00% of its market value against a 30% limit.
• Oracle was already over the line on September 30 at 38.29%. Its 6.6% fall in market value pushed the ratio 2.71 points higher with no new borrowing.
• Procter & Gamble moved the other way: its share price rose, and its debt ratio fell from 9.70% to 8.84%.
• Purifying only the interest share of dividends comes to $4.74 for the year across the portfolio. Purifying each company’s full interest income per share comes to $65.03.
• The method you follow changes the amount far more than any single stock does, so settle it with a scholar before you give.

Building the holdings check

The six stocks and share counts are a sample portfolio I set up for this article. Swap in your own and the prompts work the same way.

• Microsoft (MSFT): 20 shares
• NVIDIA (NVDA): 50 shares
• Meta Platforms (META): 15 shares
• Johnson & Johnson (JNJ): 25 shares
• Procter & Gamble (PG): 30 shares
• Oracle (ORCL): 40 shares

The three AAOIFI tests in plain words

AAOIFI Shari’ah Standard №21 sets three financial limits, each one a share of something the company has or earns:

• Debt, under 30%: how much interest-bearing debt the company carries compared with its market value. Claude divides short-term plus long-term debt by market value.
• Cash and deposits, under 30%: how much it keeps in cash and interest-bearing investments compared with its market value. Claude divides cash and short-term investments by market value.
• Interest income, under 5%: how much of its revenue comes from interest. Claude divides the last 12 months of interest income by revenue over the same 12 months.

Lease obligations, which sit on their own line in the FMP balance sheet, aren’t counted as debt, and all cash is treated as if it earns interest. Both choices follow the most common reading, but your scholar may treat them differently.

What each result means

• Compliant: passes all three tests
• Not compliant: fails at least one test
• Close to the line: passes, but debt or cash is 25% or more of market value, so a price drop could tip it over
• Check needed: a number FMP returns is zero or missing where the company clearly has cash, so Claude can’t finish the test

These labels are my way of sorting results for a personal check, separate from any accounting materiality standard.

Two ways to work out purification

Even a compliant company earns some interest, so part of what it pays you is considered impure. Investors and scholars use two common methods:

• Method A, dividend share: dividends you received × (interest income ÷ revenue). It applies to dividends only, so a stock that pays none needs nothing.
• Method B, per-share income: (interest income ÷ shares outstanding) × shares you hold. It covers your share of the company’s whole interest income, paid out or not.

Method A matches the dividend-only approach many screening apps use. Method B follows the per-share reading of AAOIFI’s standard and divides by weighted-average shares outstanding. Claude calculates both and leaves the choice to you.

If an FMP call fails, Claude tries once more. It never fills a gap with an estimate or a number from another website.

FMP data inputs and access through Claude MCP

Claude reads five kinds of FMP data. The company figures come from US filings, which FMP takes from SEC EDGAR.

• company → profile-symbol (symbol): industry and description for the business check in Table 1
• statements → balance-sheet-statement (symbol, period=quarter, limit=2): date, filingDate, shortTermDebt, longTermDebt and cashAndShortTermInvestments for Table 1
• statements → income-statements-ttm (symbol, limit=2): date, revenue, interestIncome and weightedAverageShsOut for Tables 1 and 2
• company → historical-market-cap (symbol, from_date=2026-09-30, to_date=2026-10-16): date and marketCap for Table 1
• calendar → dividends-company (symbol, limit=5): paymentDate and dividend for Table 2

All amounts are in US dollars. Company figures are in millions; your dividends and purification amounts are in dollars and cents. The balance sheet is the latest quarter filed by October 16, 2026, and the 12-month income figures end on that same quarter. Filings made after that date are left out on purpose. Dividends come from FMP’s dividend history and count if they were paid between October 17, 2025 and October 16, 2026.

Setting up FMP in Claude

1. Sign up at FMP and copy your API key from the dashboard.
2. In Claude, go to Settings → Connectors, add a custom connector and paste the server address from the FMP MCP server guide with your key.
3. Open a new chat, switch on the FMP connector, and paste Prompt 1.

Running the check in Claude

There are two prompts. The first tells you whether each stock still passes, which you might run every month or after a sharp fall. The second works out purification, which most investors only need once a year, so it reuses Prompt 1’s numbers instead of fetching them again.

Both prompts were run on October 17, 2026, using the October 16 close from FMP’s market cap history.

Prompt 1: Is each holding still compliant?

This prompt checks the business first, using the industry and description from each company’s FMP profile, then the three AAOIFI tests at two dates, so you can see whether the recent drop moved anything.

Objective: Check whether six stocks pass AAOIFI Shari'ah Standard No. 21 and show how their debt ratio changed between two dates. Explain results in plain language for a retail investor.

Holdings: MSFT, NVDA, META, JNJ, PG, ORCL.

Data rules:
- company → profile-symbol for each stock.
- statements → balance-sheet-statement, period=quarter, limit=2. Use the latest row filed on or before 2026-10-16.
- statements → income-statements-ttm, limit=2. Use the row whose date matches the balance sheet date.
- company → historical-market-cap, from_date=2026-09-30, to_date=2026-10-16. Use the 2026-09-30 and 2026-10-16 values.
- Retry a failed call once. Never estimate a value or use another source.

Steps:
1. Business check: flag the stock if its industry is a bank, insurer, alcohol, tobacco or gambling business, or its description mentions alcoholic, tobacco, casino, gambling, pork or adult entertainment.
2. Debt = shortTermDebt + longTermDebt. Debt ratio = debt ÷ market cap, on both dates.
3. Cash ratio = cashAndShortTermInvestments ÷ Oct 16 market cap.
4. Interest share = interestIncome ÷ revenue. If interestIncome is 0 while the company holds cash, mark Check needed.
5. Result: Not compliant if any test is 30% or more (5% or more for interest share), or the business check flags it. Close to the line if debt or cash ratio is 25% or more. Otherwise Compliant.

Output: Table 1. Holdings Compliance Check: Stock | Debt / market value (Sep 30) | Debt / market value (Oct 16) | Cash / market value (Oct 16) | Interest income / revenue | Result | What it means. Percentages to two decimals. Keep "What it means" to one short sentence.

Limits: Do not say whether a stock is halal. Report only what the AAOIFI financial tests show.

What Prompt 1 returned

Claude made 24 calls with no failures. None of the six companies was flagged by the business check, so every result below comes from the financial tests.

Table 1. Holdings Compliance Check

Oracle is the only holding that fails, and it failed before the drop too. What the selloff did was make it worse: Oracle’s debt stayed at $155.93 billion while its market value fell from $407.23 billion to $380.33 billion. The other five are so far from the limits that a crash of more than 70% would be needed to bring any of them to 30%.

Prompt 2: How much should I purify?

The second prompt adds your share counts and the dividends paid over the past year, then applies both purification methods side by side.

Objective: Calculate dividend purification for six holdings using two methods.

Holdings and shares: MSFT 20, NVDA 50, META 15, JNJ 25, PG 30, ORCL 40.

Data rules:
- Use interestIncome, revenue and weightedAverageShsOut from the same TTM rows used in Prompt 1.
- calendar → dividends-company, limit=5. Count only dividends with paymentDate from 2025-10-17 to 2026-10-16.
- Retry a failed call once. Never estimate a value.

Steps:
1. Dividends per share received = sum of dividend for qualifying payments.
2. Dividends received = dividends per share × shares held.
3. Method A = dividends received × (interestIncome ÷ revenue).
4. Method B = (interestIncome ÷ weightedAverageShsOut) × shares held.
5. Calculate before rounding; show dollars to two decimals. Add a total row.

Output: Table 2. Purification for the Past 12 Months: Stock | Dividends per share received | Dividends received | Interest income / revenue | Method A purification | Interest income per share | Method B purification.

Limits: Do not recommend a method. Note any holding marked Not compliant in Table 1.

What Prompt 2 returned

Claude made six dividend calls with no failures. Every holding paid four dividends in the window.

Table 2. Purification for the Past 12 Months (US dollars)

Oracle is marked Not compliant in Table 1, so its row shows what purification would be for the period, not a sign that holding it is acceptable.

The two methods differ by a factor of almost 14. Method A only cleans the slice of each dividend that came from interest. Method B assigns you a share of the company’s entire interest income, including the large part it kept instead of paying out. The gap is widest for Meta and NVIDIA, which earn billions in interest but pay small dividends.

Reading your results

For this portfolio, one decision matters: Oracle. It fails the debt test by 11 points, and the gap grew during the selloff. Ask your scholar whether to sell it, how quickly, and what to do with any gain.

The other five holdings don’t need action. Their ratios are low enough that day-to-day price moves won’t change the result, so a check every quarter, after each company reports, is enough.

On purification, the first question is which method you follow. At $4.74 under Method A and $65.03 under Method B, settle the choice before you calculate anything else. If you hold a stock that pays no dividend, Method A gives zero and Method B still gives an amount.

When to ask a scholar

• A holding shows Not compliant or Close to the line
• Claude returns Check needed because interest income is missing
• You aren’t sure whether your fund or advisor uses Method A or Method B
• A company sells something your scholar considers impermissible, even if it passed the business check

Making this a regular habit

Save both prompts and change three things each time you run them: the stock list, your share counts and the dates. Pick dates after the latest earnings season so the balance sheet figures are fresh, and keep the market value date close to the day you run it.

If you’d like to look further into a company before buying, FMP’s Insights articles walk through other checks you can run in Claude the same way, including a long-term compounder screen for finding quality companies to research.

Is Your Stock Still Shariah-Compliant After the Selloff? was originally published in Predict on Medium, where people are continuing the conversation by highlighting and responding to this story.

Note: The APIs discussed in this article are covered for research and educational purposes only. Nothing here should be considered investment advice or a recommendation to buy, sell, or trade any financial asset.

Most MCP rankings compare connectors. How many tools, how many setup steps, which chat apps they plug into. That matters for about a day. After that, all you care about is whether the numbers coming back are right.

An MCP server doesn’t create data. It’s a doorway into somebody’s database. If the company behind it licenses prices from the exchange, your agent gets exchange-grade quotes. If it scrapes them, your agent inherits every gap and stale tick, and the model will repeat them with the same confidence it uses for everything else.

So for this list I looked past the connectors at the companies running them: where their data comes from, how far back it goes, how many markets it covers and how fresh it is. I’ve worked with financial data for eight years, and these are the six providers I’d put behind an AI agent.

How I ranked the providers

Each provider was judged on the data first and the connector second:

• Where the data comes from. Exchange licenses and direct regulatory filings rank above aggregated or scraped feeds.
• Coverage. Asset classes, markets and the number of symbols an agent can ask about.
• History. Backtests, multi-year comparisons and valuation work all need depth.
• Freshness. Realtime, delayed or end of day, and how clearly the provider says which is which.
• An official MCP server. Built and maintained by the provider itself. Community wrappers didn’t qualify.

Pricing isn’t part of the ranking, because plans change often. Every fact below comes from the providers themselves, checked in October 2026.

1. Alpha Vantage

Alpha Vantage is first because of where its data comes from. The company lists official data licenses from NASDAQ, the London Stock Exchange, OPRA, S&P Dow Jones Indices and Cboe. When your agent pulls a quote, an options chain or an index level, it traces back to the exchange or index provider that produced it.

That matters more with agents than with dashboards. A human analyst might notice a strange print and check it. A model usually won’t. Starting from licensed data removes a whole class of errors before they reach the chat. The NASDAQ license also matters once your agent moves from end-of-day prices to realtime or 15-minute delayed US quotes.

Coverage and depth

• Equities: 100,000+ symbols worldwide, with realtime data and 20+ years of daily history
• Options: realtime and historical US options through REALTIME_OPTIONS and HISTORICAL_OPTIONS, on OPRA-compliant feeds with Greeks and implied volatility
• Indices: historical data for the S&P 500, Nasdaq Composite, Dow and VIX through INDEX_DATA
• Funds: prices for 4,600+ ETFs with profiles and holdings, plus mutual funds
• FX and crypto: 500+ physical and digital currencies
• Commodities: metals, energy and agriculture, with new gold and silver endpoints just added
• Analysis layers: company fundamentals, economic indicators, 50+ technical indicators such as RSI, MACD and Bollinger Bands, and the NEWS_SENTIMENT feed

All of it sits behind one API key: 130+ data functions across 10+ asset classes. For an agent, one provider covering stocks, options and macro inputs means fewer places for two sources to disagree.

The MCP server

Alpha Vantage was one of the first market data providers to ship an official hosted MCP server that exposes the full API to Claude, ChatGPT, Cursor and other MCP clients. You can limit it to the tool categories your agent needs, which helps the model choose the right function. A local version is published on PyPI for teams that prefer to self-host.

The company behind it

Alpha Vantage is backed by Y Combinator, based in Boston, and has served market data since 2017. It says more than a million developers use its APIs.

Best for: agents that need licensed market data across stocks, options, ETFs, currencies and commodities from a single source.

2. Financial Modeling Prep (FMP)

If Alpha Vantage is strongest on market data, FMP is strongest on company data. Its fundamentals for US companies come straight from SEC EDGAR, and it carries financial statements for 55,904 companies. For an agent doing valuation or earnings work, that’s the difference between reading the filing and reading someone’s summary of it.

Coverage and depth

• Securities: 70,000+ across 60+ stock exchanges, plus 4,500+ cryptocurrencies, 1,500+ forex pairs and 40 commodities
• History: up to 30+ years for most endpoints, including daily prices and fundamentals
• Research data: 10+ years of earnings reports and analyst estimates, and earnings call transcripts for 8,000+ US-listed companies
• Ownership: 13F institutional holdings, insider trades and congressional trading disclosures
• Ratings: analyst ratings and price targets from TipRanks

Freshness and transparency

FMP is open about timing. Financials usually arrive from the 8-K within 24 hours of an earnings release, then fill out when the full 10-Q lands. It also publishes the formulas behind its calculated metrics, including its DCF, Piotroski score and owner earnings. When an agent hands you a valuation built on FMP numbers, you can check exactly how each input was derived.

The MCP server

The official FMP MCP server wraps every REST endpoint as an AI tool, with strict parameter definitions so the model knows which ticker, date or period each call needs. It runs as a hosted URL in Claude, Cursor and other MCP clients.

Best for: fundamental analysis, valuation and equity research built on filing-level data.

3. FinancialData.Net

FinancialData.Net is unusually open about its own data. It names its sources, the SEC and FINRA among them, and its pricing page carries a full dataset table: what each dataset covers, how far back it goes and how often it updates. That’s exactly what you want to know before trusting an agent’s answer.

Coverage and depth

• Symbols: 252,000+ tickers across 20+ exchanges, spanning stocks, ETFs, options, futures, OTC, indices, crypto, forex and mutual funds
• Fundamentals: income statements, balance sheets, cash flows and ratios since 2010, including international statements
• Prices: end-of-day and intraday history since 2015, plus realtime prices
• Long series: corporate actions since 1964 and economic indicators going back to 1854
• Ownership and alternative data: insider transactions, Senate and House trading, institutional holdings, ETF and mutual fund holdings, ESG scores and short interest

One detail matters a lot for agent work. The API returns standard identifiers, including CUSIP, ISIN, FIGI, CIK and LEI. When an agent has to match a company across filings, holdings and prices, clean identifiers prevent the wrong-entity mix-ups that are hard to spot in a chat answer.

Reliability

The company states 99.9% uptime and runs a public status page. It says more than 1,000 professionals and institutions use the data.

The MCP server

The FinancialData.Net MCP server maps every documented endpoint to a tool, so income-statements becomes getIncomeStatements and etf-prices becomes getEtfPrices. It works with ChatGPT, Claude, Cursor and custom agents, and comes with the provider's Professional and Enterprise plans.

Best for: multi-asset research that needs clear coverage dates, long economic history and reliable identifiers.

4. EODHD

EODHD has the widest geographic reach here. Its MCP page lists 150,000+ tickers on 70+ exchanges, with stocks, ETFs, mutual funds, indices, bonds, forex and crypto in one API. If your agent covers London, Frankfurt, Sydney or Tokyo as often as New York, start here.

How the data is built

EODHD is upfront about its method. Its site says pricing data is aggregated from more than 100 sources, including OTC and trading platforms, using a VWAP method, rather than taken from exchange feeds directly. That approach is how one provider stretches across so many markets, and it suits research, screening and backtesting across regions. For order-level precision on a single US stock, a provider licensed by the exchange is the closer match.

Coverage and depth

• History: 30+ years for major US exchanges and 15 to 20 years for most European and Asian markets
• Beyond prices: fundamentals, news and sentiment, technicals, US options, US Treasury rates, ESG and macro indicators

The MCP server

EODHD’s server is one of the most carefully built for agents. It has 72 read-only tools across 15 categories and ships 100+ documentation pages as MCP resources, so the model can read how an endpoint works without spending an API call. It can resolve “Apple” or an ISIN to the right ticker, and if a request needs data outside your plan, it returns the exact permission error instead of a vague failure. You can connect with an API key or OAuth 2.1.

The company behind it, Unicorn Data Services, is based in Lyon, France, and has run the service since 2017.

Best for: cross-market research and long price histories outside the US.

5. Tradier

Tradier is the only regulated broker on this list, and that changes what “data provider” means. Its brokerage arm, Tradier Brokerage, is a member of FINRA and SIPC, and its futures arm is an NFA member. Your agent reads quotes and options chains through the same API Tradier’s customers use to place real trades.

What the data covers

Tradier is built around US stocks and options. Your agent gets realtime and delayed quotes, full options chains with strikes, expirations and Greeks, historical prices and market hours. It’s narrower than the providers above, but deep where active traders spend their time.

The MCP server

The hosted Tradier MCP server launched in December 2025 and covers the full Tradier API: market data, options chains, account balances, watchlists and order placement. Paper trading is built in, so an agent can test a strategy on simulated money before anything touches a live account. It works with ChatGPT, Claude, Cursor and Gemini, and the setup docs walk through each one.

Best for: options traders who want analysis and execution in one place, starting on paper.

6. Nasdaq Data Link

Nasdaq Data Link carries the strongest brand on this list and works differently from the rest. It’s a marketplace of financial, economic and alternative datasets built for investment professionals, and you subscribe to the datasets you need rather than one bundled API.

That makes it the place to look for data the other five don’t focus on: alternative datasets, long economic series and specialist research tables.

The MCP server

The Nasdaq Data Link MCP gives AI agents one connection to every dataset your organization subscribes to. Entitlements and access controls are enforced on every call, so the model only sees what your license covers. For a research team juggling several premium datasets, that governance is the main draw: one integration, with compliance handled at the source.

Access is aimed at existing Nasdaq data clients, who can contact Nasdaq to connect. For quick experiments with free datasets, an open-source community server built on Nasdaq’s official Python SDK covers sources such as World Bank data.

Best for: institutions that already license Nasdaq datasets and need alternative or economic data under strict access control.

Which provider fits your work

Alpha Vantage is my overall pick. Licensed market data across stocks, options, ETFs, currencies and commodities makes it the safest default to put behind an agent. FMP is a very close second, and the first one I’d reach for when the question is about a company’s financials rather than its share price.

The other four are strong choices for specific jobs:

• FinancialData.Net when you want multi-asset coverage with every dataset’s start date and update schedule written down
• EODHD when your watchlist runs across Europe, Asia and Australia
• Tradier when the agent should be able to act on an options idea, on paper first
• Nasdaq Data Link when your firm already licenses Nasdaq datasets and needs them governed inside AI tools

Many setups connect two. A market data provider paired with a fundamentals provider covers most research questions, and the agent can cross-check a number when the two disagree. Keep the combined tool count in check, though. A model handed hundreds of tools spends its effort choosing between them instead of answering you.

A year ago, getting stock data into an AI chat meant pasting CSVs or writing glue code. Today most serious data providers ship an MCP server. You can ask Claude for a company’s last eight quarters of revenue and get real numbers back, not a guess.

The hard part now is picking one. GitHub has dozens of finance MCP servers. Many are weekend projects that wrap a free API and fall over on the third call. A smaller group is official, maintained by the data provider itself, and connects with one URL.

I’ve worked with financial data for eight years. For this list I only looked at servers that can handle real stock analysis: pulling statements, prices, options and macro data into an agent. Six made the cut.

Quick pick

• Best overall: Alpha Vantage
• Best for fundamentals and valuation: Financial Modeling Prep (FMP)
• Best for wide coverage behind one key: FinancialData.Net
• Best for international stocks: EODHD
• Best for trading from the chat: Tradier
• Best for licensed and alternative datasets: Nasdaq Data Link

How I picked and ranked them

Every server here had to clear the same bar:

• Official or provider-backed. Community wrappers were left out, even good ones.
• Depth of stock data. Fundamentals, price history, options, technical indicators and news all count. More coverage means fewer servers to juggle.
• Setup effort. A hosted URL you paste into Claude beats a local Python install.
• A free way to try it. You should be able to test before you pay.
• Documentation. Clear tool lists and example prompts save hours.

The ranking reflects which server I’d tell most people to start with. Pricing below was checked against each provider’s site in October 2026.

1. Alpha Vantage

Alpha Vantage is the one I’d hand to almost anyone as the default choice for a comprehensive stock market data MCP server. It was one of the first market data providers to ship an official MCP server, and it puts its whole API behind it: more than 130 data functions covering 10+ asset classes including stocks, ETFs, mutual funds, indices, fundamentals, options, forex, crypto, commodities, economic indicators, news, and more.

The company is a NASDAQ-licensed data provider, which matters once you move from end-of-day prices to realtime or 15-minute delayed quotes.

What’s new and worth knowing

Options are the big addition. Premium keys get realtime and historical US options chains through the REALTIME_OPTIONS and HISTORICAL_OPTIONS functions, with Greeks and implied volatility included. For anyone screening covered calls or checking IV before earnings, that removes the need for a second provider.

There’s also INDEX_DATA, which returns historical data for major indices like the S&P 500, Nasdaq Composite, Dow and VIX. Add 50+ built-in technical indicators (RSI, MACD, Bollinger Bands) and the NEWS_SENTIMENT feed, and one server covers most of what a stock analysis agent asks for.

How to connect

The hosted server needs no install. Paste this into Claude, Cursor or any MCP client:

https://mcp.alphavantage.co/mcp?apikey=YOUR_API_KEY

By default it exposes every tool. If your agent only needs equities and options, you can trim the tool list with a categories parameter, for example &categories=core_stock_apis,options_data_apis. Fewer tools means the model picks the right one more often. Prefer running it locally? The official Python package starts with uvx alphavantage-mcp.

Pricing

The free key allows 25 requests a day across most datasets, and verified open-source or educational projects get unlimited requests. Premium plans have no daily cap:

Annual billing works out to roughly ten months’ cost. Personal users unlock realtime data through a quick entitlement step in Alpha Vantage’s Alpha X Terminal after subscribing.

Best for: Anyone who wants prices, fundamentals, options and technicals from one official server, with a free tier to try it first.

Try this prompt: “Pull NVDA’s last four quarterly earnings, then show the options chain for the nearest monthly expiry with implied volatility.”

Alpha Vantage MCP Server

2. Financial Modeling Prep (FMP)

If your analysis lives in financial statements, FMP is the deepest server on this list. Its official MCP server wraps every REST endpoint FMP offers as an AI tool. That includes income statements, balance sheets and cash flows, plus TTM versions, growth rates, key metrics, ratios, revenue segments and as-reported filings.

Where it pulls ahead for analysts is the ready-made valuation work. There are DCF and levered DCF endpoints, a custom DCF calculator, Piotroski and Altman scores, owner earnings and enterprise values. Ask Claude for a valuation and it can start from FMP’s numbers instead of building every input from scratch.

On the research side, you get earnings call transcripts, 13F institutional holdings, insider trades, Senate and House trading disclosures, SEC filings, analyst estimates and price targets. Analyst ratings and price targets from TipRanks are a newer addition to the catalog.

How to connect

In Claude, open Settings → Connectors → Add custom connector and paste:

https://financialmodelingprep.com/mcp?apikey=YOUR_FMP_API_KEY

The same URL works in Cursor, Cloudflare’s Workers AI Playground, or a Python agent through FastMCP. Each tool call counts against your normal API limit, so there’s no separate MCP plan to buy.

Pricing

FMP’s plans for personal use:

Best for: fundamental analysis, valuation and equity research where you need statements, ratios and filings in one place.

Try this prompt: “Compare Apple and Microsoft’s free cash flow margin and ROIC over the last five years, then run FMP’s DCF for both.”

AI Agent (MCP Server) - FMP

3. FinancialData.Net

FinancialData.Net is the youngest provider here, and it moves fast. The API’s first release was in December 2024. The changelog since then reads like a steady build-out: options chains and Greeks, futures, international financial statements, ETF holdings, forex, economic data and a news feed.

The official MCP server launched on February 16, 2026, and every endpoint in the docs maps to a tool. income-statements becomes getIncomeStatements, etf-prices becomes getEtfPrices, and so on. That one-to-one mapping makes the tool list easy to reason about when your agent picks the wrong one.

Coverage is broad for the price. The pricing page lists 252,000+ tickers across 20+ exchanges, fundamentals back to 2010, and economic indicator history going back to 1854. Insider trades, Senate and House trading, institutional holdings, ESG scores and short interest are all in the catalog.

Two recent additions stand out. July 2026 brought an Analyst Consensus endpoint and a full Economic Data group. September 2026 added Universal Query, an endpoint for custom data pulls.

How to connect

Add this URL to Claude, Cursor or your own agent:

https://financialdata.net/mcp?key=API_KEY

To confirm it worked, ask your AI to list the tools from the financial data server.

Pricing

MCP access comes with the Professional plan at $149/month ($1,299/year) and the Enterprise plan at $299/month ($2,599/year). Professional includes realtime data, the Excel add-in and internal commercial use. A free tier with 300 requests a day lets you test the REST API before upgrading.

Best for: teams that want wide asset coverage, including options, futures and international fundamentals, behind one hosted URL.

Try this prompt: “Check NVIDIA’s trailing twelve-month return against the S&P 500 and calculate the alpha.”

4. EODHD

EODHD wins on global reach. Its MCP server covers 150,000+ tickers on 70+ exchanges with 30+ years of history, through 72 read-only tools in 15 categories. If your analysis touches European, Asian or Australian stocks, this is the one to check first.

Two details show real care for agent users. The server ships 100+ pages of EODHD’s documentation as MCP resources, so the model can read how an endpoint works without spending an API call. It also comes with three prompt templates, analyze_stock, compare_stocks and market_overview, that chain several tools for you.

How to connect

There are two versions. v1 uses your API key and suits ChatGPT, Claude Code and custom agents:

https://mcp.eodhd.com/v1/mcp?apikey=YOUR_API_KEY

v2 uses OAuth 2.1, which keeps the key out of the URL and works well in Claude Desktop: https://mcp.eodhd.com/v2/mcp.

Pricing

The MCP server is included on every plan, free tier too. The free plan gives 20 calls a day. Paid plans start at $19.99/month for end-of-day data, $29.99 adds intraday and technicals, $59.99 covers fundamentals, and the All-in-One plan is $99.99/month. Annual billing saves two months. US options with Greeks and IV for 6,000 stocks are a $29.99/month add-on.

Best for: international stock coverage and long price histories.

5. Tradier

Tradier is the only broker on this list, and that changes what your agent can do. Its hosted MCP server, launched in December 2025, gives you realtime quotes and options chains with Greeks. It also lets the agent check positions and P&L, then place orders, from single stocks up to multi-leg spreads and bracket orders.

For options traders that’s a tight loop. Ask for SPY’s chain, have the model pick strikes for a put spread, then submit it to a paper account to see how it behaves. Paper trading endpoints come with every account, so nothing touches real money until you switch it on.

How to connect

Generate a personal access token in your Tradier account, then add this URL to Claude, Cursor or Gemini CLI:

https://mcp.tradier.com/mcp?api_key=YOUR_TOKEN&paper_trading=true

Set paper_trading=false only when you're ready to trade live.

Pricing

Any Tradier account works with the MCP server, including the free Lite plan.

Best for: options traders who want analysis and execution in the same chat.

6. Nasdaq Data Link

Nasdaq Data Link is a marketplace more than a single API. It hosts financial, economic and alternative datasets from Nasdaq and third-party publishers, and you subscribe to the ones you need.

The Nasdaq Data Link MCP gives your agent one governed connection to every dataset you’re subscribed to. Entitlements are checked on every call, so the model only sees what your licence covers. For a research team juggling several premium datasets, that’s one integration instead of five.

It’s also the right place to look for data the other five don’t carry, like alternative datasets and long economic series. If you just want to experiment, an open-source community server built on Nasdaq’s official Python SDK works with free datasets such as World Bank indicators.

How to connect

Existing Nasdaq Data Link clients can follow the connection guide linked from the MCP product page, or contact Nasdaq to get set up.

Pricing

There’s no single plan. Cost depends on which datasets you subscribe to, and some are free for personal use.

Best for: research teams already licensing Nasdaq datasets, or anyone who needs alternative and economic data beyond standard prices and fundamentals.

Which stock market MCP server should you choose?

For most stock market workflows, Alpha Vantage and Financial Modeling Prep (FMP) are the two best MCP servers in 2026, and each one is strongest in a different area, with Alpha Vantage edging ahead as the best all-rounder.

Alpha Vantage wins on range. Prices, fundamentals, technical indicators, options with Greeks, news sentiment and macro data all sit with one provider. An analyst checking earnings, a developer building a screener and a trader watching implied volatility can all work from the same key, and the plans scale from a free test all the way to realtime options.

FMP is the server I’d trust most with a company’s numbers. Statements and prices go back 30+ years on the higher plans, ratios and key metrics come ready to use, and DCF valuations, analyst estimates, earnings transcripts and 13F holdings are a single prompt away. For valuation, screening and peer comparisons, you won’t find a deeper source on this list.

FinancialData.Net is a great fit if you want options, futures, international statements and institutional data behind one key, and its catalog keeps growing every few months. EODHD is the one to reach for when your watchlist goes beyond the US, with 70+ exchanges and 30+ years of history.

Tradier suits options traders who want analysis and execution in the same chat. And for teams that work with licensed, economic or alternative datasets, Nasdaq Data Link is the natural choice.

If you mainly need those connection steps, my list of MCP servers for finance data covers the setup URL and sign-in flow for each one.

Frequently asked questions

What is the best MCP server for stock market analysis?
For most people, Alpha Vantage. Prices, fundamentals, options, technical indicators and news all come through one server. If your work is mostly financial statements and valuation, FMP is the better pick.

Is there a free MCP server for stock market data?
Yes. Alpha Vantage’s free key gives you 25 requests a day. EODHD includes its MCP server on the free plan with 20 calls a day, and Tradier works with its free Lite account.

Which MCP server has options data with Greeks?
Alpha Vantage (premium keys), Tradier and FinancialData.Net all return options chains with Greeks. EODHD offers them as a paid add-on.

Do I need to install anything?
Usually not. Most of these servers give you a URL to paste into Claude, ChatGPT or Cursor. Alpha Vantage can also run locally with uvx alphavantage-mcp.

Can an AI agent place trades through MCP?
On this list, only Tradier can. Keep paper_trading=true until you’re ready to trade with real money.

Read More From Author

• Top 10 Financial APIs in 2026
• Top Stock Market APIs You Must Be Aware of (Updated 2026)
• Top MCP Servers for Finance Data

Best MCP Servers for Stock Market Analysis in 2026 was originally published in Level Up Coding on Medium, where people are continuing the conversation by highlighting and responding to this story.

Claude can explain crypto, summarize market narratives, and reason through investment questions surprisingly well.

But crypto moves too quickly for static knowledge to be enough.

Prices change around the clock. News breaks continuously. Market sentiment shifts. An altcoin can move sharply within hours, while an AI model relying only on previously learned information may have no idea what just happened.

So I wanted to test what changes when Claude no longer has to rely on static knowledge for the financial facts behind its analysis.

For this experiment, I connected Claude directly to Financial Modeling Prep through its MCP server and used NEAR as the test asset.

I then gave Claude a series of practical research tasks and evaluated how well it could retrieve, connect, and summarize the evidence.

The experiment was intentionally simple:

• Check what NEAR is doing in the market right now
• Put the recent price movement into historical context
• Find relevant recent developments and news
• Combine the evidence into a concise research brief

No large crypto screen. No comparison across ten tokens. No attempt to build an automated trading bot.

The goal was simply to see whether connecting Claude to live financial data through MCP could make crypto research faster, more current, and more useful.

In other words, instead of asking Claude to remember what it knows about NEAR, we are asking it to retrieve the current evidence first and reason from that evidence.

That is the experiment.

Let’s see how it performs.

The Setup: Claude + FMP MCP

The setup is simple:

Claude → FMP MCP Server → Live Crypto Data

Financial Modeling Prep supports MCP, which means Claude can discover and query FMP’s available financial tools directly instead of requiring us to manually call APIs, parse JSON, and paste the results into a prompt. FMP’s crypto coverage includes real-time quotes, historical price data, and cryptocurrency news, which is enough for the experiment in this article.

To connect FMP in Claude:

1. Open Settings → Connectors
2. Select Add custom connector
3. Paste the FMP MCP URL:

https://financialmodelingprep.com/mcp?apikey=YOUR_FMP_API_KEY

4. Replace YOUR_FMP_API_KEY with your active FMP API key

5. Start a new Claude conversation

Once connected, Claude can discover the FMP tools available through the MCP server and use them when responding to the prompts we send.

For this article, we will keep the experiment focused on NEAR and test four things:

current market data → recent price context → relevant news → a concise research brief

AI Agent (MCP Server) - FMP

The point is not to build a trading system. It is to see how much more useful Claude becomes when the facts behind the analysis come from a current financial data source.

Test 1: What Is NEAR Doing Right Now?

Let’s start with the most basic research task: understanding what NEAR is doing in the market right now.

Normally, this would mean opening a crypto-price website, checking the current price, looking at the recent move, switching to a historical chart, and manually putting the numbers into context.

With FMP connected through MCP, Claude can retrieve that information directly.

For this first test, I asked Claude to:

• Confirm the correct NEAR/USD symbol available through FMP
• Retrieve the latest price, percentage change, trading volume, and daily range
• Calculate the 7-day and 30-day price change
• Compare the latest move with NEAR’s recent 30-day price behavior

The goal here is not to predict NEAR’s next move. It is simply to see whether Claude can retrieve current market data, perform the calculations, and turn the raw numbers into useful market context without any manual data collection.

Using only data available through the connected Financial Modeling Prep MCP server, analyze the current market position of NEAR Protocol.

1. First, identify and confirm the correct NEAR/USD cryptocurrency symbol available in FMP.
2. Retrieve the latest available crypto quote for NEAR and report:
   - Price
   - Latest percentage change
   - Trading volume
   - Day high and low, if available
   - Timestamp or data date
3. Retrieve historical daily prices and calculate:
   - 7-day price change
   - 30-day price change
   - Highest and lowest closing price over the last 30 days
4. Briefly explain whether the latest move appears small, moderate, or unusually large relative to NEAR's recent 30-day price behavior.

Use only FMP data. Do not use web search or outside sources.

Do not give a buy, sell, long, or short recommendation.

Keep the response concise and return:
- A small market snapshot table
- 3–5 bullet points explaining the recent price context

What Test 1 Showed

The first test already shows the difference between asking Claude about a cryptocurrency and letting Claude retrieve the market data behind its answer.

Claude correctly identified NEARUSD as the FMP symbol and combined the latest quote with historical prices to provide both short-term and 30-day context.

At the time of the run, NEAR was trading around $4.94, down 8.18% on the latest quote. Looking only at that daily move could make the decline appear unusually severe.

The historical data changed the picture.

NEAR was still up roughly 16% over seven days and more than 160% over the previous 30 days according to the FMP data used by Claude. The historical series also showed several double-digit daily moves during the month.

That allowed Claude to place the latest decline in context: it was a relatively large daily move, but not outside the range of volatility NEAR had already experienced during the recent rally.

The useful part here is not the price itself. Any crypto website can show that.

The value comes from Claude being able to retrieve the current quote, pull the relevant history, calculate the changes, and summarize the market context in one interaction — without us manually collecting and pasting the numbers first.

Test 2: What Changed Around NEAR?

The first test established what NEAR was doing in the market.

The next question is more useful:

What changed around NEAR while this move was happening?

Crypto news can be noisy. A single token may appear in exchange announcements, ecosystem updates, partnership stories, market commentary, and speculative headlines within the same week.

Rather than reading everything manually, I asked Claude to use FMP’s crypto news data to identify the developments most relevant to NEAR’s recent move.

The goal was not to force an explanation for the rally.

Instead, I wanted Claude to:

• Retrieve recent NEAR-related news from FMP
• Remove duplicate or low-value headlines
• Identify the developments that appear most relevant
• Place those developments on the same timeline as NEAR’s recent price movement
• Clearly separate confirmed facts from possible interpretation

This is an important distinction. A headline appearing near a price move does not prove that it caused the move.

Using only data available through the connected Financial Modeling Prep MCP server, review recent developments related to NEAR Protocol.

1. Confirm the NEAR/USD symbol used by FMP.
2. Retrieve NEAR-related cryptocurrency news from the most recent 14 days.
3. Remove obvious duplicate stories and low-value mentions.
4. Select up to 5 developments that appear most relevant to NEAR itself.
5. For each development, report:
   - Date
   - Headline
   - Source
   - Short summary
   - Why it may matter for NEAR
6. Compare the timing of these developments with NEAR's recent price movement from FMP historical data.

Important rules:
- Use only FMP data.
- Do not claim that any news item caused the price move unless the evidence clearly establishes that relationship.
- Clearly distinguish confirmed facts from interpretation.
- Do not give buy, sell, long, or short recommendations.

Return:
- A compact table of the most relevant developments
- A short timeline summary explaining what changed around NEAR during the recent move

What Test 2 Showed

This test was more useful than simply asking Claude to summarize crypto news.

Claude retrieved a large number of NEAR-related headlines, recognized that many of them were repetitive price alerts, and reduced the noise to a smaller set of developments that were actually relevant to the project.

Several meaningful events appeared during the same period:

• NEAR Intents crossed roughly $30 billion in cumulative volume
• Ondo brought tokenized U.S. stocks and ETFs to near.com and NEAR Intents
• Hyperliquid added NEAR/USDC spot trading
• Bitwise’s proposed NEAR ETF moved through important SEC and NYSE Arca registration steps
• NEAR introduced private limit orders for cross-chain assets

The timeline also showed why this kind of analysis needs caution.

The strongest phase of NEAR’s price advance occurred around the same time that ETF-related coverage intensified, but that timing alone does not prove the ETF news caused the rally. Several ecosystem developments appeared within the same few days, while the broader crypto market was also moving.

Claude handled this reasonably well by separating temporal overlap from confirmed causation.

That is one of the more useful things an MCP-connected workflow can do: retrieve the relevant headlines, remove obvious noise, align events with price data, and leave the final interpretation open where the evidence is not strong enough.

Test 3: Can Claude Turn the Evidence Into a Research View?

By this point, Claude has already retrieved NEAR’s recent price behavior and the developments reported around the same period.

The next test is harder:

Can it turn that evidence into a useful research view?

This is where an AI assistant becomes more interesting than a simple data-retrieval interface.

Instead of asking for another list of prices or headlines, I asked Claude to combine the available FMP evidence into three things:

• The strongest factors supporting the current NEAR narrative
• The main risks or reasons for caution
• The developments worth monitoring next

The goal is not to generate a price target or decide whether NEAR should be bought or sold.

It is to see whether Claude can move from retrieving information to organizing an investment-research question around evidence.

Using only data available through the connected Financial Modeling Prep MCP server, build a concise research view for NEAR Protocol based on the latest available market data and recent developments.

Use the recent price history, crypto news, and other relevant FMP evidence available for NEAR.

Structure the response into four sections:

### 1. Current Market Context
Summarize NEAR's recent price trend and volatility using the latest available FMP data.

### 2. Factors Supporting the Current Narrative
Identify up to 4 developments that could reasonably support investor interest in NEAR.

For each one:
- State the confirmed fact
- Explain why it may matter
- Do not assume it caused the price move

### 3. Risks and Reasons for Caution
Identify up to 4 risks or uncertainties supported by the available evidence.

Consider:
- Recent price volatility
- Whether the rally has been unusually fast
- Uncertainty around the financial impact of recent developments
- Whether important claims remain unverified or incomplete

Do not introduce risks that are not supported by the FMP evidence.

### 4. What to Watch Next
List up to 5 specific things that would be useful to monitor over the coming days or weeks.

Focus on observable items such as:
- Follow-up ETF developments
- Trading or liquidity changes
- New ecosystem announcements
- Changes in price behavior
- New NEAR-related news

Important rules:
- Use only FMP MCP data.
- Clearly separate confirmed facts from interpretation.
- Do not provide a price target.
- Do not predict that NEAR will rise or fall.
- Do not provide buy, sell, long, or short recommendations.
- Keep the output concise and evidence-based.

What Test 3 Showed

This was the point where Claude moved beyond data retrieval.

Instead of returning another collection of prices and headlines, it combined the evidence into a structured research view: what currently supports the NEAR narrative, what remains uncertain, and what deserves monitoring next.

The distinction between fact and interpretation was especially useful.

For example, Claude could confirm that the Bitwise NEAR ETF had moved through important SEC and NYSE Arca registration steps. But it also recognized that the available evidence did not yet provide a confirmed first trading date.

Similarly, the NEAR Intents volume milestone and Ondo integration were real developments, but the data available in this experiment did not establish their direct financial impact on the NEAR ecosystem.

That is a much more useful output than simply labeling every development “bullish.”

Claude was able to organize the research into three buckets:

What we know → What it might mean → What we still need to verify

And importantly, it did this without turning the result into a price prediction or trading recommendation.

For me, this is where the MCP connection becomes most useful. FMP supplies the current evidence, while Claude helps turn that evidence into a research process.

Want to Try It Yourself? Use This Prompt

After running the three tests separately, the pattern is fairly clear.

Claude works best when the prompt does three things:

1. Tells it exactly which asset to research
2. Restricts the analysis to current FMP MCP data
3. Separates confirmed facts from interpretation

If you want to repeat the same workflow for another supported cryptocurrency, you can use a prompt like this:

Using only data available through the connected Financial Modeling Prep MCP server, research [CRYPTO ASSET].

Confirm the correct USD symbol available in FMP.

Retrieve the latest quote and summarize:

Current price

Latest percentage change

Trading volume

Day high and low, if available

Retrieve historical prices and calculate:

7-day change

30-day change

30-day high and low

Retrieve relevant crypto news from the most recent 14 days.

Remove duplicate or low-value headlines and identify the most important developments.

Compare the timing of those developments with the recent price movement, but do not assume causation.

Build a concise research view with:

Current market context

Factors supporting the current narrative

Risks and uncertainties

What to watch next

Important rules:

Use only FMP MCP data.

Clearly separate confirmed facts from interpretation.

Do not invent missing information.

Do not provide a price target.

Do not give buy, sell, long, or short recommendations.

Keep the response concise and evidence-based.

The asset can be changed without rewriting the entire workflow.

For example, replacing NEAR with another supported cryptocurrency lets Claude repeat the same process: retrieve current data, add historical context, filter recent developments, and organize the evidence into a research view.

The prompt is not designed to predict the next breakout.

It is designed to make the first stage of crypto research faster and more structured.

AI Agent (MCP Server) - FMP

Conclusion

This experiment started with a simple question: what changes when Claude can access current financial data instead of relying only on what it already knows?

Using NEAR as the test asset, Claude was able to move through three stages of research:

retrieve the market data → investigate recent developments → synthesize the evidence into a research view

The interesting part was not any single number or headline. It was the workflow itself.

With FMP connected through MCP, Claude could retrieve the relevant data, filter out noise, add historical context, and organize the findings through natural-language prompts.

That makes MCP especially interesting for financial research.

Instead of manually collecting data first and asking the model to analyze it afterward, the research process becomes much more direct:

ask the question → retrieve the evidence → analyze the context

And NEAR was only one example.

The same approach can be extended across other supported cryptocurrencies, equities, financial statements, earnings data, news, and other datasets available through FMP.

As MCP-connected tools continue to expand, the more interesting question may no longer be simply what Claude knows.

It may be:

What data can Claude reach when you ask the question?

Curious how FMP compares with other providers? I tested six servers side by side in Best MCP Servers for Stock Market Analysis in 2026.

I Gave Claude Live Stock Market Data Through MCP — Here’s What It Could Do in 10 Minutes was originally published in Predict on Medium, where people are continuing the conversation by highlighting and responding to this story.

A weekly investment committee meeting should focus on decisions and exceptions, not on manually assembling information. Yet for many family offices, a significant part of the preparation still involves checking portfolio movements, reviewing recent company developments, looking for newly reported financial information, identifying upcoming earnings events, and deciding which holdings actually deserve discussion.

This is a good fit for a structured Claude + Financial Modeling Prep (FMP) MCP workflow. Instead of asking Claude to generate a generic portfolio summary, the family office can define a repeatable review process that tells Claude what evidence to collect, what period to review, how to prioritize developments, and when an issue should be escalated for analyst attention.

In this article, we will build a practical weekly investment committee briefing workflow using Claude connected directly to FMP through its MCP server. The demonstration will use a small portfolio sample, but the process will be designed as a reusable operating framework. Claude will review recent price performance, check for relevant company news and newly available financial information, identify upcoming earnings catalysts, and convert the evidence into a concise portfolio review and committee agenda.

The goal is not to let Claude make investment decisions. Instead, the workflow automates much of the repetitive preparation that happens before those decisions are made. Holdings with little material change can remain out of the discussion queue, while developments that require interpretation, conflicting evidence, or additional investigation can be routed to the analyst as Review Required.

By separating data collection, prioritization, analyst review, and recurring execution, the same framework can move beyond a one-time prompt and become part of a family office’s weekly investment committee process.

What We Are Building

The objective is to create a repeatable weekly review that helps a family office determine which portfolio holdings deserve investment committee attention.

Rather than asking Claude to deeply analyze every company, the workflow first performs a focused review across a defined set of signals available through the FMP MCP server. Claude collects the relevant evidence, evaluates whether anything material has changed, and converts the findings into a prioritized committee briefing.

The workflow follows this sequence:

Portfolio Holdings → Weekly Review Window → FMP Data Retrieval → Development and Catalyst Review → Priority Classification → Investment Committee Briefing → Analyst Review

For this demonstration, we will intentionally keep the portfolio small. The same framework can later be expanded to a larger portfolio without changing the underlying review logic.

These inputs are supported by datasets available through Financial Modeling Prep. FMP’s MCP server allows Claude to dynamically retrieve market prices, company fundamentals, earnings information, filings, and news rather than requiring each dataset to be manually downloaded and combined.

One distinction is important for the fundamental review. Financial statements do not change simply because a new weekly review is being run. The workflow should therefore check whether new financial information became available during the review period, rather than forcing a week-over-week fundamental comparison when no new filing or report exists.

The result is a two-stage process. Holdings with no meaningful new evidence remain out of the main discussion queue, while potentially important developments are surfaced for analyst review and, when necessary, added to the investment committee agenda.

Accessing FMP Data Through Claude MCP

The weekly briefing needs several types of evidence, but they do not need to be downloaded and combined manually. Financial Modeling Prep’s MCP server allows Claude to retrieve FMP datasets directly while running the review.

An active FMP API key is required. In Claude, open:

Settings → Connectors → Add custom connector

Use the following remote MCP server URL:

https://financialmodelingprep.com/mcp?apikey=YOUR_FMP_API_KEY

Replace YOUR_FMP_API_KEY with an active FMP API key. Once the connector is enabled in a new Claude conversation, Claude can discover the available FMP tools and call them as required by the analysis.

For this workflow, each dataset has a specific role rather than being included simply because it is available.

The workflow should preserve the dates attached to these records. A news article published this week, a financial statement from the latest completed quarter, and an earnings event scheduled for next week represent different types of evidence and should not be treated as though they share the same observation date.

Financial data also requires a slightly different review rule. Claude should not compare financial statements merely because the weekly job has run again. It should first determine whether a new financial statement became available during the defined review period. If no new statement exists, the briefing can record that no new fundamental reporting evidence was identified rather than manufacturing a weekly fundamental change.

Before running the complete workflow, it is useful to validate the MCP connection with one portfolio company. Confirm that Claude can retrieve a recent historical price observation, company-specific news, the latest available financial statement, and the next available earnings-calendar record. This small check verifies the connector, API access, symbol, and required datasets before the full briefing is generated.

Once those inputs are available, the next step is to define the rules that determine which developments should remain informational and which should be escalated for investment committee attention.

Defining the Weekly Investment Committee Review Rules

Before Claude reviews the portfolio, the family office needs to define what should actually qualify for investment committee attention. Without explicit rules, the model may surface too many routine developments or apply inconsistent judgment from one weekly review to the next.

For this demonstration, we will use three review states:

No Action: No development identified that requires additional attention in the current review cycle.

Monitor: A noteworthy development exists, but the available evidence does not yet justify escalation to the investment committee.

Review Required: The holding requires analyst attention because the development may be material, evidence is conflicting or incomplete, or an upcoming event warrants preparation before the committee meeting.

The following rules provide a reproducible starting point for the demo:

These thresholds are intentionally simple so the workflow can be reproduced in an article demonstration. They are not intended to represent a universal investment committee policy. A real family office could adjust them based on portfolio volatility, mandate, position size, liquidity requirements, or internal risk limits.

Treat Rules as Triggers, Not Investment Conclusions

A threshold breach should determine what receives attention, not whether a security should be bought or sold.

For example, a 6% weekly price increase would trigger Review Required, but Claude should not conclude that the holding is overvalued or recommend taking profit. Its job is to investigate whether FMP data contains a development that may help explain the move and surface the issue for analyst review.

The same principle applies to news. Claude should give higher attention to developments such as:

• Management guidance changes
• Earnings or financial reporting developments
• Mergers, acquisitions, or major divestitures
• Material regulatory or legal events
• Capital raises or significant financing activity
• CEO, CFO, or other major leadership changes
• Major product, customer, or strategic announcements

Routine commentary, duplicated headlines, or market-wide stories that do not provide company-specific information should not automatically enter the committee agenda.

Use the Highest Applicable Review State

More than one signal may apply to the same holding. The workflow should assign the highest review state triggered by the available evidence.

For example, a company may have moved only 2% during the week, which would normally remain No Action. However, if it is scheduled to report earnings in three days, the holding should still be classified as Review Required.

Likewise, a large price move without a clearly supported explanation should not be dismissed simply because Claude cannot find a matching headline. The absence of sufficient evidence is itself a reason for analyst review.

This produces a simple operating principle:

Retrieve the evidence → apply the predefined triggers → assign the highest applicable review state → escalate only the holdings that need additional attention.

With the review policy defined, Claude can now collect the required FMP evidence using the same rules every week rather than deciding from scratch what is important.

How to Follow This Workflow

The next sections explain the logic behind the weekly investment committee review — how the portfolio and review window are defined, what evidence Claude should retrieve through the FMP MCP server, how holdings are prioritized, and what the final briefing should contain.

These sections are meant to show how the workflow is designed. You do not need to run a separate Claude prompt at every step.

Instead, we will combine the review dates, evidence requirements, escalation rules, and output format into a single Claude + FMP MCP prompt later in the article and run the complete workflow once.

If you want to understand or customize the methodology, follow the steps in sequence. If you mainly want to see the implementation, you can move directly to the Final Claude + FMP MCP Prompt section and then review the actual briefing generated from it.

Step 1: Define the Portfolio and Review Window

The weekly workflow should begin with a fixed portfolio input and explicit review dates. These inputs become the reference point for every subsequent FMP query.

For the demonstration, we will use two liquid U.S. equities:

The small sample keeps the Claude run manageable while still showing how the workflow would operate across companies from different sectors.

The review should also use two separate time windows:

• Historical review window: Previous 7 calendar days
• Forward catalyst window: Next 14 calendar days

The historical window determines which price movements, news, press releases, and newly available financial information belong to the current weekly briefing. The forward window is used to identify upcoming earnings events that may require preparation.

Using explicit dates is important. If Claude is simply asked to retrieve “recent developments,” different runs may cover different periods and make one week’s briefing difficult to compare with the next.

For each weekly run, the workflow should therefore establish three dates before retrieving any company data:

Review start date → As-of date → Catalyst end date

For example, if the briefing is prepared after the final trading session of the week, the as-of date should correspond to that completed market period rather than an incomplete future session.

Why the Portfolio Input Should Stay Separate

In a real family office implementation, the portfolio list may eventually come from a portfolio management system, database, spreadsheet, or internal API. Claude should therefore treat the list of holdings as an input rather than embedding company names permanently into the analytical logic.

The article uses only two symbols, but the same structure can later accept a larger list without changing the review methodology.

At this stage, Claude’s task is deliberately narrow: validate the symbols, establish the review dates, and confirm the inputs that will be used by the remaining workflow. No risk classification or investment committee prioritization should happen yet.

Step 2: Retrieve the Weekly Evidence From FMP

Once the portfolio and review dates are fixed, the next step is to build the evidence set that will support the weekly briefing.

At this stage, Claude should act primarily as a data collector and organizer. It should retrieve the relevant information through the FMP MCP server without yet deciding whether a holding is No Action, Monitor, or Review Required.

For each portfolio company, the workflow collects four evidence groups.

1. Measure the Market Move

Retrieve historical end-of-day prices covering the review window and calculate the change between the first available trading-day close and the latest completed trading-day close.

This gives the workflow a reproducible market-movement measure instead of relying on an intraday quote that may change while the briefing is being prepared.

The calculation is:

7-day price move (%) = ((Ending Close / Starting Close) — 1) × 100

The underlying start date, end date, and closing prices should also be retained so an analyst can verify the calculation.

2. Collect Recent Company Developments

Next, retrieve company-specific stock news and press releases published within the historical review window.

The goal is not to copy every headline into the briefing. Claude should first preserve the available evidence, including:

• Publication date
• Headline
• Source
• Related symbol
• Source URL, when available
• Whether the item came from company-issued information or external news

Duplicate or substantially similar stories can be consolidated later, but they should not be interpreted as material at the retrieval stage.

3. Check for Newly Available Financial Information

Claude should then check the latest available financial statements and determine whether new reporting information became available within the weekly review window.

If nothing new was reported, the workflow should explicitly record:

No new financial reporting identified during the review window.

This is preferable to repeatedly analyzing an older income statement or balance sheet as though it represented a new weekly development.

If new reporting is available, Claude should retain the reporting period and relevant dates before any interpretation takes place.

4. Check the Forward Earnings Calendar

Finally, Claude should search the defined 14-day forward window for a scheduled earnings event.

If an earnings event exists, retain the expected date and the relevant calendar fields available from FMP. If no event is found in the period, record that explicitly.

At the end of this stage, each holding should have a compact evidence packet containing its market move, recent company developments, any newly available financial reporting, and upcoming earnings information.

No committee priority should be assigned yet. Separating evidence collection from classification makes it easier to verify the underlying data before Claude begins making judgments about what deserves attention.

Step 3: Convert the Evidence Into Review Priorities

Once the weekly evidence has been collected, Claude can apply the predefined review rules to determine which holdings deserve additional attention.

This stage is intentionally separate from data retrieval. Claude should classify each holding using only the evidence gathered from FMP and the rules defined earlier in the workflow.

For every holding, evaluate four signals:

• 7-day price movement
• Recent company news or press releases
• Newly available financial reporting
• Upcoming earnings within the 14-day catalyst window

The final review status should be based on the highest applicable trigger.

Apply the Quantitative Triggers First

The deterministic rules are straightforward:

Price movement

• Absolute move below 3% → No Action
• Absolute move from 3% to below 5% → Monitor
• Absolute move of 5% or more → Review Required

Upcoming earnings

• No earnings within 14 days → No Action
• Earnings in 8–14 days → Monitor
• Earnings within 7 days → Review Required

These rules provide a consistent baseline and reduce unnecessary model judgment.

Then Review Qualitative Developments

News, press releases, and newly available financial information require more interpretation.

A development can remain No Action when it is routine or unlikely to affect the investment thesis.

It can move to Monitor when it is relevant enough to track but does not yet appear to require committee discussion.

It should move to Review Required when the evidence indicates a potentially material development or when additional analyst judgment is needed.

Examples include:

• A change in management guidance
• A significant acquisition, divestiture, or financing event
• A major regulatory or legal development
• A material earnings or financial reporting surprise
• A significant management change
• A major strategic, customer, or product announcement

Claude should not infer materiality solely from an attention-grabbing headline. The classification should be tied to the evidence available through FMP.

Escalate Uncertainty Instead of Guessing

The workflow should also treat uncertainty as a review condition.

For example, a holding may experience a large weekly price move while FMP shows no clearly related company development. In that case, Claude should not invent an explanation.

Instead, it should classify the holding as:

Review Required — material market move without sufficient supporting evidence.

Similarly, conflicting information, missing dates, or incomplete records should be surfaced for analyst verification rather than resolved through unsupported assumptions.

Assign One Final Status Per Holding

A holding may trigger more than one rule during the same weekly review.

For example:

• Price move: 2.4% → No Action
• Earnings in 10 days → Monitor
• Material management announcement → Review Required

The final status should therefore be:

Review Required

This highest-trigger approach gives the investment committee a simple prioritization mechanism while still preserving the underlying reasons for escalation.

At the end of this stage, Claude should produce a concise review queue showing:

The purpose is not to decide what the family office should do with the holding. It is to decide what deserves human attention before the investment committee meeting.

Step 4: Generate the Weekly Investment Committee Briefing

After the review status has been assigned, Claude can convert the evidence into a concise investment committee briefing.

The objective is not to create a long research report for every holding. Instead, the briefing should answer two practical questions:

1. What changed across the portfolio this week?
2. What actually deserves investment committee discussion?

To keep the output compact, the workflow produces two final tables.

Weekly Portfolio Review

The first table gives the committee a portfolio-level snapshot.

The Key Development column should contain only the most relevant evidence identified during the review period. If no meaningful company development or new financial reporting is found, Claude should state that clearly instead of filling the column with routine news.

The Status column uses the previously defined classification:

No Action / Monitor / Review Required

This table helps the analyst quickly understand which holdings changed enough to warrant attention.

Investment Committee Agenda

The second table should include only issues that deserve further discussion.

This is the most important output of the workflow.

A holding should not automatically appear in the investment committee agenda simply because Claude found news about it. It should be included only when the review rules identify a meaningful trigger.

For example, the agenda might surface:

• A holding with a large weekly price move and no clearly supported catalyst
• A company reporting earnings within the next few days
• A new financial report containing a potentially significant development
• A major company announcement requiring interpretation
• Conflicting or incomplete evidence that needs analyst verification

Keep Evidence Separate From Interpretation

For every item included in the agenda, Claude should preserve the underlying FMP evidence.

Where available, this should include:

• Relevant date
• Data type
• Headline or reported event
• Source
• Source URL
• Trigger responsible for the review status

This allows the analyst to verify the issue before it reaches the committee.

Claude can summarize why an item deserves attention, but it should not convert that interpretation into an investment recommendation.

For example:

Acceptable:
“Review Required because the stock moved 6.2% during the review window and the available FMP evidence does not provide a clearly supported company-specific explanation.”

Not acceptable:
“The stock is overextended and should be sold.”

Keep the Briefing Exception-Driven

The workflow becomes more useful when the final briefing is selective.

If a holding has:

• No significant market move
• No material new company development
• No new financial reporting
• No near-term earnings catalyst

then it does not need to occupy valuable committee time.

The goal is therefore not to summarize every holding equally. It is to help the family office filter routine information and direct analyst attention toward exceptions, catalysts, and unresolved issues.

This makes the weekly briefing a prioritization tool rather than another portfolio report.

Putting It All Together: The Final Claude + FMP MCP Prompt

The review logic is now defined: what evidence to collect, how to prioritize each holding, and what the final committee briefing should contain.

Instead of running these stages separately, we can combine them into a single Claude prompt and let the FMP MCP server execute the complete weekly review.

You are helping prepare a weekly investment committee briefing for a family office.

Use **only Financial Modeling Prep data available through the connected FMP MCP server**.

### Portfolio

Analyze only:

* AAPL
* JPM

Keep the workflow compact. Do not expand the portfolio, run a broad market screen, or perform unnecessary follow-up searches.

---

## Step 1 — Establish the Review Period

1. Identify the **latest completed U.S. trading day available in FMP** and use it as the as-of date.
2. Set the historical review window to the **previous 7 calendar days ending on the as-of date**.
3. Set the forward catalyst window to the **next 14 calendar days after the as-of date**.
4. Use explicit dates throughout the analysis.

---

## Step 2 — Collect the Weekly Evidence

For each company, retrieve only the evidence required below.

### A. Market Movement

Retrieve historical end-of-day prices covering the 7-day review window.

Use:

* The first available trading-day closing price within the review window
* The latest completed trading-day closing price

Calculate:

**7-Day Price Move (%) = ((Ending Close / Starting Close) - 1) × 100**

Preserve:

* Starting date
* Starting close
* Ending date
* Ending close
* Calculated percentage move

Do not use an incomplete intraday price.

### B. Company Developments

Retrieve company-specific:

* Stock news
* Press releases

published during the historical review window.

Keep a maximum of **5 distinct relevant developments per company**.

For each retained development, preserve:

* Publication date
* Headline
* Source
* Source URL, if returned by FMP
* Whether it is external news or a company press release

Remove obvious duplicate coverage.

Do not infer that a news event caused the stock-price movement unless the available evidence clearly supports that connection.

### C. Newly Available Financial Information

Check the latest available financial statements and determine whether **new financial reporting became available during the historical review window**.

If yes, preserve:

* Statement type
* Fiscal period
* Reporting/publication date available through FMP
* The specific reported development relevant to the weekly review

If not, state:

**No new financial reporting identified during the review window.**

Do not treat an older financial statement as a new weekly development.

### D. Upcoming Earnings

Check the FMP earnings calendar for the 14-day forward catalyst window.

If an earnings event exists, preserve:

* Scheduled earnings date
* Relevant calendar information returned by FMP

If none exists, state:

**No earnings event identified within the 14-day forward window.**

---

## Step 3 — Apply the Review Rules

Use these demo rules exactly.

### Price-Movement Trigger

Use the **absolute value** of the 7-day price move.

* Below 3% → **No Action**
* 3% to below 5% → **Monitor**
* 5% or greater → **Review Required**

### Upcoming Earnings Trigger

Based on calendar days from the as-of date:

* No earnings within 14 days → **No Action**
* Earnings in 8–14 days → **Monitor**
* Earnings within 7 days → **Review Required**

### News / Press Release Trigger

* Routine or clearly immaterial development → **No Action**
* Relevant development worth tracking → **Monitor**
* Potentially material development requiring analyst interpretation → **Review Required**

Examples that may justify closer review include:

* Management guidance changes
* Material earnings or reporting developments
* Significant acquisition or divestiture
* Major financing or capital-raising activity
* Significant regulatory or legal development
* CEO, CFO, or other major leadership change
* Major strategic, customer, or product announcement

Do not classify an item as material solely because the headline sounds important.

### Financial-Reporting Trigger

* No new reporting → **No Action**
* New reporting with information worth monitoring → **Monitor**
* New information requiring deeper analyst assessment → **Review Required**

### Evidence-Quality Trigger

Assign **Review Required** when:

* Important evidence is conflicting
* A significant issue cannot be adequately verified
* A large market move lacks sufficient supporting company-specific evidence
* Required FMP data is materially incomplete

Do not fill evidence gaps with assumptions.

### Final Holding Status

Assign **one final status per holding**:

* No Action
* Monitor
* Review Required

Use the **highest applicable status triggered by any review signal**.

These statuses represent **review priority only**, not buy, sell, or hold recommendations.

---

## Step 4 — Produce the Weekly Investment Committee Briefing

Return only the following sections.

### 1. Review Configuration

Provide:

* As-of date
* Historical review window
* Forward catalyst window

Keep this to a few lines.

### 2. Weekly Portfolio Review

Return this table:

| Holding | 7-Day Move | Key Development | Upcoming Catalyst | Status |
| ------- | ---------: | --------------- | ----------------- | ------ |

Rules:

* One row per company
* Keep Key Development concise
* If no meaningful development exists, say so
* Do not fill the table with routine headlines

### 3. Investment Committee Agenda

Return this table:

| Priority | Holding | Issue for Discussion | Supporting Evidence | Analyst Follow-Up |
| -------- | ------- | -------------------- | ------------------- | ----------------- |

Include only holdings/issues that are **Monitor or Review Required** and are useful for committee preparation.

If neither holding deserves committee discussion, state that clearly instead of manufacturing an agenda item.

Use:

* **High** priority for Review Required
* **Medium** priority for Monitor

### 4. Supporting Evidence

After the two tables, provide a compact evidence list for AAPL and JPM containing only evidence actually used in the classifications.

Include:

* Date
* Evidence type
* Relevant value/headline/event
* FMP source URL when available

Also show the starting and ending prices used in each 7-day return calculation.

---

## Evidence and Interpretation Rules

* Use only FMP MCP data.
* Do not use general web search or outside sources.
* Do not invent missing data.
* Do not provide buy, sell, hold, valuation, or portfolio-allocation recommendations.
* Do not claim that news caused a price move unless the evidence establishes that connection.
* Keep qualitative explanations conservative.
* Put uncertain interpretation into **Analyst Follow-Up**.
* Clearly distinguish publication dates, reporting periods, and scheduled future dates.
* If an FMP field is unavailable, mark it unavailable rather than estimating it.
* Do not repeatedly retry unavailable data or expand the analysis beyond the requested scope.

The goal is to identify **what deserves human attention before the investment committee meeting**, not to automate the investment decision itself.

Results: What the Weekly Review Surfaced

Running the consolidated workflow through Claude with the FMP MCP server produced two different outcomes across the sample portfolio.

AAPL was classified as Monitor, while JPM was escalated to Review Required.

Weekly Portfolio Review

The classifications illustrate why the workflow evaluates more than price movement alone.

AAPL: Small Price Move, but a Development Worth Monitoring

AAPL moved only +0.62% during the review period, well below the 3% price threshold that would independently trigger monitoring.

There was also no earnings event within the 14-day forward window.

However, the FMP evidence surfaced two company developments during the week. Apple renewed its global patent-licensing agreement with Qualcomm, while the Apple Pay antitrust litigation advanced to class certification.

Neither development, based on the evidence retrieved, justified an investment conclusion. The Qualcomm agreement did not provide enough information to determine its financial impact, while class certification represents a procedural development rather than a finding that Apple violated antitrust law.

The workflow therefore placed AAPL in Monitor rather than escalating it directly to Review Required.

JPM: Escalation Came From Context, Not the Price Threshold

JPM declined 2.55% during the review window.

Importantly, that move alone would still fall under No Action according to the predefined price rule because its absolute change remained below 3%.

The escalation came from the qualitative evidence instead.

During the same review period, J.P. Morgan Asset Management announced a $20 billion strategic partnership with the Qatar Investment Authority, while JPM shares weakened alongside broader pressure in financial stocks.

The available evidence did not establish that the partnership caused the subsequent price move, nor did it provide enough support to attribute the decline to one specific factor.

Rather than inventing a causal explanation, the workflow classified JPM as Review Required and routed the unresolved question to an analyst.

Investment Committee Agenda

The resulting agenda focuses only on the issues that deserve additional attention.

This is the key difference between a conventional portfolio summary and an exception-driven investment committee workflow.

Claude does not simply report everything that happened. It filters the evidence using predefined rules and surfaces the smaller set of issues that may deserve analyst or committee attention.

The result is a briefing designed around what changed, what remains uncertain, and what requires human review.

Analyst Review Triggers

The weekly briefing is designed to reduce manual preparation, not eliminate analyst judgment.

Claude can collect FMP evidence, apply predefined thresholds, and prioritize holdings. However, certain situations should always be routed to a human analyst before they become part of an investment committee discussion.

When a holding is classified as Review Required, Claude can perform a focused follow-up using the same FMP MCP connection. The follow-up should investigate only the issue that triggered the escalation rather than rebuilding the entire company review.

For example, if the escalation came from an unexplained price move, the follow-up can focus on company news, press releases, newly available financial information, and upcoming events that may provide additional context. If the available evidence still does not establish a clear explanation, the unresolved question should remain with the analyst rather than being filled with an assumption.

1. A Large Move Without a Supported Explanation

If a holding experiences a meaningful price move but the available FMP evidence does not clearly explain it, Claude should not infer a cause.

The holding should remain Review Required, with the unresolved question passed to the analyst.

This is especially important when several possible explanations exist, such as company news, sector movement, macro developments, or changing market expectations.

2. Qualitative Developments With Unclear Financial Impact

Company announcements can be important without immediately revealing their economic significance.

Examples include:

• Strategic partnerships
• Major customer agreements
• Regulatory developments
• Litigation
• Management changes
• Product announcements
• Acquisitions or divestitures

If the available evidence does not establish the likely financial impact, Claude should describe the development conservatively and ask the analyst to investigate further.

3. Conflicting Evidence

Different FMP records may point in different directions.

For example, a company may announce a positive strategic development while the stock weakens, or a press release may appear constructive while external reporting highlights additional risks.

Claude should preserve the disagreement rather than forcing the evidence into a single narrative.

4. Missing or Incomplete Data

A missing field should not be replaced with an estimate.

Analyst review is appropriate when:

• A relevant source date is unclear
• A financial-impact figure is unavailable
• An expected event cannot be verified
• Supporting company-specific evidence is insufficient
• A requested FMP field is missing

This prevents unsupported assumptions from entering the investment committee briefing.

5. Materiality Requires Portfolio Context

FMP and Claude can identify developments at the company level, but materiality may depend on information that is not part of the MCP workflow.

A family office analyst may also need to consider:

• Position size
• Portfolio concentration
• Investment mandate
• Tax considerations
• Liquidity requirements
• Existing hedges
• Private information from managers or counterparties

A development that is minor for one portfolio could be significant for another.

Keep the Human Review Focused

The purpose of these triggers is not to send every holding back to an analyst.

Instead, the workflow should automate routine evidence gathering and allow human attention to concentrate on:

uncertainty, conflicting evidence, potentially material developments, and portfolio-specific interpretation.

That division of work keeps the weekly process efficient while ensuring that Claude remains a decision-support tool rather than an autonomous investment decision-maker.

Automate and Scale the Weekly Briefing

The demonstration uses AAPL and JPM in a manually initiated Claude session, but the same workflow can be adapted into a recurring pre-committee process.

Schedule the Weekly Review

A family office can run the workflow on a fixed schedule before each investment committee meeting.

For example, if the committee meets on Monday morning, the review could run after the final U.S. trading session on Friday or early Monday before the meeting.

The recurring process becomes:

Scheduled Trigger → Load Portfolio → Set Review Dates → Query FMP → Apply Review Rules → Generate Briefing → Analyst Review

Load the Portfolio Dynamically

In the article demonstration, AAPL and JPM are written directly into the prompt.

In a broader implementation, the holdings should come from a maintained source such as:

• A portfolio management system
• Internal database
• Approved spreadsheet or CSV
• Portfolio API

This allows the analytical instructions to remain stable while the portfolio changes over time.

Screen First, Investigate Exceptions

For a larger portfolio, the same review logic can be applied in manageable batches.

Each holding is first classified as:

No Action / Monitor / Review Required

The initial pass should remain lightweight. Only holdings classified as Review Required need deeper follow-up.

This creates a practical two-layer model:

Layer 1 — Portfolio Screening
Identify which holdings deserve attention.

Layer 2 — Exception Analysis
Investigate only the issues that require additional context.

This prevents the family office from performing the same level of research on every holding every week.

Preserve the Weekly History

Each briefing should be saved with its as-of date rather than replacing the previous week’s output.

Over time, Claude can compare the current briefing with earlier reviews and identify:

• New issues
• Persistent issues
• Escalated issues
• De-escalated issues
• Resolved issues

This helps the investment committee track how portfolio concerns evolve rather than treating every weekly briefing as an isolated snapshot.

Add Portfolio Context as the Workflow Grows

The article focuses on company-level evidence available through FMP, but a production workflow can later incorporate internal portfolio information such as:

• Position size
• Sector exposure
• Concentration
• Liquidity requirements
• Internal risk limits

The same company development may have a very different impact depending on how important that holding is to the portfolio.

The two-company demonstration therefore represents the smallest version of a broader operating model:

collect evidence → classify holdings → investigate exceptions → consolidate committee priorities → preserve the history for the next review.

Conclusion

Preparing for a weekly investment committee meeting does not have to begin with manually checking prices, scanning company news, reviewing financial updates, and building an agenda from scratch.

With Claude connected to the FMP MCP server, a family office can turn that work into a repeatable review process. The workflow can establish the review period, collect the relevant evidence, apply predefined escalation rules, and surface only the holdings that deserve additional analyst or committee attention.

The AAPL and JPM demonstration also shows why the process should be exception-driven. A small price move may still deserve monitoring when a meaningful company development appears, while a larger discussion may be required when the available evidence is incomplete or does not clearly explain what happened.

The goal is not to automate the investment decision. It is to automate the repetitive preparation around that decision so analysts can spend more time on interpretation, portfolio context, and the questions that actually require judgment.

For family offices, the practical value comes from making this process consistent and repeatable: collect the evidence, prioritize the exceptions, investigate what matters, and arrive at the investment committee meeting with a focused agenda.

How Family Offices Can Build a Weekly Investment Committee Briefing With Claude + FMP MCP was originally published in Predict on Medium, where people are continuing the conversation by highlighting and responding to this story.