How Family Offices Can Use the FMP MCP Server for Investment Research

A three-person research desk at a family office does not necessarily need three market-data terminals. It may need structured access to specific public-market datasets, such as Form 13F filings, insider transactions, equity prices, and analyst expectations, along with a practical way to incorporate that data into repeatable workflows.

That matters if you are running research for a family office, managing a small fund with a lean team, or helping an outsourced CIO practice build its technology stack. The problem is often not a lack of information. It is the time spent moving between disconnected tools and turning the results into something an investment committee can use.

The Workaround Many Small Investment Teams Use

The pattern is familiar. One analyst pulls institutional ownership data from one source. Another checks insider transactions somewhere else. Someone exports a watchlist to a spreadsheet before every investment committee meeting because none of the tools share a common workflow.

Nobody built this process on purpose. It accumulated one subscription and one workaround at a time: a market-data feed here, an ownership-tracking tool there, and another source for analyst estimates or economic data.

The cost is not limited to the subscriptions. It is also the time spent assembling and reconciling information before the analysis can begin. For a lean investment team, that is time not spent examining the company, challenging assumptions, or preparing for the questions the committee is likely to ask.

Why API-First Access Matters for Smaller Teams

Large terminals and enterprise platforms are designed to support broad institutional workflows. A family office with two analysts may still need institutional ownership data, fundamentals, historical pricing, analyst estimates, and market calendars, but it may not need the execution, communications, and specialized functions bundled into a full terminal environment.

API-first providers make it possible to assemble a more focused research stack. The distinction is important: financial data APIs can replace parts of a terminal-based workflow, but they do not replace every terminal feature or every internal system.

Financial Modeling Prep provides structured datasets across financial statements, historical market prices, Form 13F filings, insider activity, analyst estimates, earnings calendars, and other public-market categories. FMP's MCP Server adds a natural-language access layer for supported AI clients.

How FMP's MCP Server Fits

The Model Context Protocol, or MCP, gives an AI client a standard way to call tools supplied by an external data source. FMP's MCP Server makes supported FMP endpoints available to clients such as Claude without requiring the user to write a separate API wrapper for each interactive request.

For a family office or small fund, several parts of FMP's coverage may be particularly useful:

• Form 13F data: Review reported institutional ownership and compare completed filing periods.
• Insider transactions: Examine reported activity by company insiders and open the associated filings when links are returned.
• Historical and current market data: Review prices for public equities and other supported asset classes.
• Analyst estimates and price targets: Add market expectations to a company review without treating them as FMP recommendations.
• Financial statements and company information: Bring operating performance and issuer context into a defined research task.

The value is not that Claude replaces the analyst. It is that the analyst can request a limited set of FMP data in plain language and ask for it in a consistent table or summary. The output still needs to be checked against the returned fields, dates, and source context.

Traditional Fragmented Process	FMP MCP-Assisted Process
Ownership, insider activity, and prices reviewed in separate tools	Supported FMP datasets can be requested through one connected client
Data copied manually into a shared document or spreadsheet	Returned data can be organized into a defined table or summary
Each analyst develops a different research format	The team can reuse the same scope and output structure
Missing data may be discovered late in the process	The request can require missing fields or periods to be identified

This does not mean FMP replaces every specialized research product a fund may use. It means the baseline public-market data layer can be brought into one interactive research environment when the relevant dataset is available under the connected FMP plan.

Set Up the FMP MCP Connection

Getting started requires an active FMP API key and an MCP-compatible client. In Claude, the connection can be added through the Connectors settings when custom connectors are available for the account or workspace.

1. Get an FMP API key through the FMP dashboard.
2. Add the API key to FMP's hosted MCP connection URL.
3. In Claude, open Settings, select Connectors, and choose Add custom connector.
4. Paste the connection URL into the Remote MCP Server URL field, give the connector a name, and select Add.

FMP MCP Connection URL

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

Keep the API key inside the connector configuration. Do not place it in prompts, screenshots, shared documents, or exported chats. Each MCP request counts toward the API limits associated with the FMP account, and dataset availability varies by plan.

Start with one small request and confirm that Claude can see the expected FMP tools. Treat that first request as a controlled sandbox test by checking the returned field names, observation dates, and available datasets before designing a recurring workflow.

Workflow 1: Review Institutional Ownership Before an IC Meeting

Before presenting a public company to an investment committee, an analyst may want to see how reported institutional ownership changed between two completed quarters. The task is useful, but the dates need to be handled carefully. Form 13F filings reflect quarter-end positions and are reported later, so they should not be described as current holdings.

Reliable quarter-over-quarter 13F ownership analysis depends on using consistent reporting periods and change conventions. FMP's Positions Summary API accepts a symbol, year, and quarter. Keep the first request to one company and two explicit reporting periods.

Prompt suggestion: Ask Claude to retrieve the positions summary for the same ticker in two specified year-quarter periods and place the fields available in both periods into a side-by-side table. The request can ask for the number of investors, shares held, total investment value, ownership percentage, and change fields when they are returned. It should also require Claude to identify a missing period or field rather than selecting a substitute.

The finished comparison should show the symbol, both year-quarter labels, the returned ownership measures, and any calculated change. The analyst should confirm the periods and recalculate at least one figure before using the table in an investment committee document.

If insider activity or analyst targets are also relevant, review them as separate follow-up tasks rather than combining everything into one large request. The Search Insider Trades API can return reported transactions and filing links, while the Price Target Consensus API provides high, low, median, and consensus targets.

Those datasets also require different interpretations. An insider sale may be an open-market transaction, an option exercise, a planned trade, or another reported event. Transaction type, ownership form, filing date, and transaction date can materially affect the interpretation when using insider-transaction data in governance analysis. An analyst target represents a third-party expectation, not intrinsic value or an FMP recommendation. Keeping the checks separate makes those distinctions easier to preserve.

Workflow 2: Check Concentration Across User-Supplied Entities

Family offices rarely manage only one entity. The same public security may appear in a trust, foundation, personal account, or investment vehicle, each with its own reporting needs.

FMP MCP does not connect to a custodian and discover those holdings. The analyst must supply the positions. FMP market data can then be used to price a limited list and calculate concentration within the information provided.

For the first pass, keep the list to no more than five publicly traded equities, use entity aliases instead of account numbers, and select one completed end-of-day valuation date. Starting with one asset type and one currency avoids the timing and conversion problems created by combining equities, crypto, forex, or private assets in the same test.

Entity Alias	Symbol	Units	Currency
Trust A	AAPL	2,500	USD
Foundation	SPY	1,200	USD
Investment LLC	MSFT	900	USD

Prompt suggestion: Ask Claude to use FMP's historical end-of-day stock prices to retrieve the closing price for each supplied symbol on the same completed valuation date. It can then calculate the value of each row, combine repeated symbols, show each position's weight within the successfully valued list, and flag positions above a threshold chosen by the user.

The output should show the requested date, actual price date, returned symbol, closing price, units, calculated value, and concentration weight. Any missing price, uncertain symbol, or unresolved currency should appear separately rather than being converted to zero or omitted silently.

This remains a public-equity concentration screen, not a complete family-office exposure report. It does not include cash, liabilities, private assets, derivatives, tax lots, beneficial ownership, fund look-through, or accounts that were not supplied. The denominator should be labeled accordingly.

Workflow 3: Prepare a Quarterly Price-Change Summary

Every quarter, someone has to turn a watchlist into a summary that highlights which securities deserve attention. A complete performance report requires position weights, cash flows, dividends, corporate actions, and a benchmark series. A simpler price-change review can still identify the largest moves before the analyst begins that deeper work.

Use no more than five public-equity symbols for the initial review. Select two completed quarter ends and one consistent price convention. If the review uses dividend-adjusted prices, use the Dividend Adjusted Price Chart API for every symbol rather than mixing adjusted and unadjusted data.

Prompt suggestion: Ask Claude to retrieve the selected end-of-day price for each ticker at the prior and current quarter ends, calculate the signed percentage change, and rank the completed results by the absolute size of the move. If a quarter end falls on a non-trading day, the output should show the actual prior trading date used. Symbols with missing prices should be listed separately and excluded from the ranking.

The result should show each ticker, requested dates, actual observation dates, prior price, current price, signed percentage change, and price basis. It should not describe the figures as portfolio returns, contribution, or performance attribution, and it should not invent a reason for any price move.

The analyst should recalculate the largest positive and negative changes and review any corporate action or ticker change before interpreting an extreme result.

When MCP Fits and When It Does Not

These workflows are best suited to a family office or small fund with a lean research team, a defined public-market task, and an analyst who understands the requested data well enough to review the result. MCP is particularly useful for interactive research, one-time comparisons, and testing a consistent output format without building a custom integration first.

The REST API is a better fit when a process must run on a schedule, cover a large universe, feed a database, or apply the same calculation automatically. A recurring production process also needs logging, retry rules, schema monitoring, and other controls outside the chat interface. Teams can move from a small API test to a repeatable workflow after the underlying data and calculation have been reviewed.

FMP MCP is less relevant when the main requirement is private-company valuation, portfolio accounting, custody reconciliation, tax reporting, order execution, or a complete total-wealth view. Those tasks require other systems, internal data, and additional controls.

Protect Sensitive Information

Do not place custody statements, account numbers, personal identifiers, wallet credentials, private-company records, or unrestricted beneficial-ownership information into an AI workspace without authorization. Use the minimum information required for the public-market task and replace real entity names with aliases when possible.

For any repeated workflow, keep a record of the supplied symbols, requested periods or dates, returned observation dates, FMP dataset used, missing fields, and analyst review. The purpose is not to create a complicated governance process for a small test. It is to preserve enough context to explain how the result was produced.

If access later expands beyond one researcher, define who owns the workflow, which outputs may be shared, and where the data can be used. A financial-data access review can help establish appropriate responsibilities and boundaries before the same connection is used across research, operations, technology, or external reporting.

Key Takeaways

• FMP MCP can bring supported public-market data into Claude without requiring a custom API wrapper for each interactive request.
• The most useful starting points are narrow, recurring tasks: comparing two Form 13F periods, pricing a small user-supplied equity list, or reviewing quarter-end price changes.
• Large research requests should be divided into separate checks. Smaller requests use fewer resources and are easier for an analyst to verify.
• MCP does not replace source review, investment judgment, portfolio accounting, or the systems required to manage private assets and sensitive account data.

Get Started

Start with one symbol, one pair of completed periods, or one short synthetic position list. Confirm that Claude can retrieve the expected FMP data, check the returned dates and fields, and recalculate the material figures before expanding the task.

The FMP MCP documentation provides the current connection steps. If the workflow grows beyond a small test, match the FMP plan to the required datasets, historical depth, refresh cadence, and number of symbols. The current FMP pricing page provides the latest information about dataset access and request capacity.

Frequently Asked Questions

Can a small team use FMP MCP without hiring a developer?

For interactive use in a supported client, adding the remote connector does not require building an API wrapper. A developer may still be needed if the team wants to automate the workflow, integrate it with internal systems, or apply production controls.

Does FMP cover asset classes beyond public equities?

FMP provides datasets for asset classes including equities, ETFs, crypto, forex, and commodities. Coverage, timing, symbol conventions, and plan access differ by dataset. The examples in this article stay focused on public equities to keep the first workflows easier to review.

Does FMP MCP replace a market-data terminal?

No. It can support defined data retrieval, comparison, and formatting tasks inside an MCP-compatible AI client. It does not reproduce every terminal function, including execution, communications, proprietary analytics, and interactive monitoring.

Can FMP MCP create a complete family-office exposure report?

Not from public-market data alone. A complete report may require cash, liabilities, private assets, derivatives, tax lots, beneficial ownership, fund look-through, and entity relationships from internal systems.

Are MCP outputs investment advice?

No. The workflows organize data and calculations for analyst review. They do not establish suitability, predict returns, explain a price move, or replace investment, legal, tax, compliance, or risk judgment.

Do MCP requests have separate usage limits?

FMP's current documentation states that requests made through the MCP Server count toward the API limits associated with the connected account. Dataset access and request capacity depend on the account's plan.