Best Finance MCP Servers in 2026

The Model Context Protocol (MCP) is an open standard that lets an AI assistant call external tools. A finance MCP server is one of those tool providers, pointed at money: it wraps a brokerage, a data feed, or a research source and hands the assistant a set of callable tools. Mechanically, the flow is simple. You connect the server to your MCP client (Claude Desktop, ChatGPT, or another host). When you ask a question, the model decides which tool to call, the server executes that call against the underlying API or brokerage, and the raw result comes back into the conversation for the model to reason over. Authentication happens at the server, not in the chat: a data server holds an API key, a brokerage server holds an account or aggregation credential, and a well-designed one stays read-only until you explicitly approve an action.

The category splits cleanly into three kinds, and the split is what this whole guide turns on:

• Brokerage and execution servers. They link a real account so the assistant can reason over what you actually own and, where the platform allows it, place trades. Alpaca, Interactive Brokers, TradeStation, SnapTrade-based servers, and Walnut’s connector live here. This is the only kind that sees your positions.
• Market-data servers. They wrap a quotes-and-fundamentals API so the assistant has current numbers to work with. Finnhub, Polygon.io, Alpha Vantage, and Financial Modeling Prep are common ones. They serve the market in general, not your account.
• Research servers. They expose filings, earnings transcripts, and sourced analysis so the assistant can fetch the actual document and cite it, rather than recalling it from training data.

A data or research server talks about the market in the abstract. A brokerage-connected one talks about your money. Both are useful; they answer different questions, and many people run more than one. If you are new to the term, the explainer on what an MCP connector is covers the basics, and MCP connector vs API vs plugin explains how it differs from a raw API.

Brokerage and execution servers are the only kind that reach a real account. They let the assistant read holdings and, where the platform allows it, place trades. Alpaca and Interactive Brokers are the developer picks (raw API access, code required); TradeStation offers similar access for its own customers; SnapTrade is the regulated aggregation layer that reaches many brokers behind one integration. To be upfront, since this is our site: Walnut’s connector is the portfolio-aware, consumer-facing option in this group, and it leads in that narrow category rather than overall.

The distinctive part of a brokerage-connected server is that the chat knows a real account. For developers, the alpaca-mcp-server and IBKR wrappers give raw programmatic control. For someone who just wants their existing broker grounded in a conversation without writing code, Walnut’s connector is read-only by default, frames each holding against the S&P 500, and routes any trade to your broker only after you approve it; Walnut is not an investment adviser. Being hosted, its connector needs no code: you add the server URL, https://walnutinvest.com/api/connector/mcp, as a custom connector in Claude, ChatGPT, or Codex (or any MCP client), then sign in with a free Walnut account and approve read-only access, and it never places trades on its own. The full per-client setup lives on the Walnut MCP connector page. To go deeper on wiring an account into an assistant, see how to connect a brokerage to an AI assistant.

Market-data servers give the assistant current numbers to reason with. They wrap a quotes-and-fundamentals API, so instead of relying on stale training data the model can call a tool for a live price, a historical series, or a company’s fundamentals. They are broad and useful, but they serve the market in general and know nothing about your account.

These are the right call when the question is about the market itself: a quote, a fundamentals check, a price history. Finnhub and Alpha Vantage are common free-tier starting points, Polygon.io is a paid pick for deep price history, and Financial Modeling Prep leans toward statements and valuation. They are the wrong call when you want the chat grounded in what you actually own, which needs a brokerage-connected server instead. Many people pair a market-data server with a portfolio connector so the assistant has both the live numbers and the holdings.

Research servers expose deeper sources: SEC filings, earnings transcripts, analyst notes, and curated news. Rather than trusting the model to recall a document, the assistant calls a tool that fetches the real filing or dataset and reasons over it with citations back to the source, which is a meaningfully safer way to research. Many are open-source community projects on GitHub that wrap public endpoints such as SEC EDGAR.

Research servers are the right call when you want depth on a company or a filing and want the assistant to cite its source. Like market-data servers, they inform your research without seeing your positions, so they complement rather than replace a brokerage-connected server.

A large share of finance MCP servers are community-built and published on GitHub. Many are the very servers named above: the alpaca-mcp-server is open source, wrappers for Interactive Brokers and SnapTrade are community projects, and market-data servers for Finnhub, Polygon.io, and Alpha Vantage often ship as inspectable repositories. They are transparent, self-hostable, and easy to fork, which makes them a natural starting point for developers wiring finance tools into an assistant themselves. The trade-off is that quality, maintenance, and reliability vary widely, most are read-only data wrappers with no portfolio grounding, and you own the setup and upkeep. A hosted connector like Walnut’s is maintained for you at the cost of the transparency of running your own.

The fastest way to choose is to name what you want the assistant to reach, then pick the server built for that. There is no overall number one; Walnut’s connector leads only in its own category (a consumer chat grounded in your real portfolio), not across the board.

• You write code and want to trade an account programmatically. Alpaca (alpaca-mcp-server) or Interactive Brokers give raw, agent-ready API access; TradeStation does the same for its own customers.
• You want one integration across many brokers. SnapTrade is the regulated aggregation layer to build on.
• You want the chat to know your real holdings without coding. Walnut’s connector reaches the brokerage you already have, read-only by default, with trades on approval.
• You want live quotes and fundamentals. A market-data server (Finnhub, Polygon.io, Alpha Vantage, or Financial Modeling Prep) feeds the assistant current prices and fundamentals to reason with.
• You want filings and sourced analysis. A research server fetches the actual document so the assistant can cite it instead of recalling it.
• You want more than one of these. Running a data server alongside a brokerage connector is common, so the assistant has both the market numbers and your positions.

Once you know whether you want account access, live data, or sourced research, a few practical filters narrow it the rest of the way:

• What does it connect to? This is the first question. A market-data server (Finnhub, Polygon.io, Alpha Vantage, FMP) or research server never sees your account; only a brokerage server (Alpaca, IBKR, TradeStation, SnapTrade-based, or Walnut’s connector) does. Pick the one that reaches what you actually need.
• Do you want to write code? Alpaca and Interactive Brokers are developer tools; Walnut’s connector is the no-code consumer path over an existing broker.
• Is it read-only by default? For anything touching your brokerage, prefer read-only-by-default access and explicit approval for any trade. Walnut’s connector reads holdings read-only and approves every trade with you.
• How does access and aggregation work? If a server connects to your money, prefer regulated aggregation (SnapTrade) over a home-rolled credential grab.
• How fresh and complete is the data? Free or public endpoints (Alpha Vantage, Finnhub free tiers) can be delayed or rate-limited. Check the underlying source before you rely on the numbers.
• Who maintains it? A hosted connector is maintained for you; a community server on GitHub is yours to keep running. Weigh the upkeep against the transparency.

There is no single best finance MCP server, because they connect to different things. To reach a real account, brokerage and execution servers are the category: Alpaca and Interactive Brokers for developers, TradeStation for its own customers, SnapTrade as the aggregation layer, and Walnut’s connector as the portfolio-aware consumer option that links your existing broker securely, read-only by default, and routes any trade only after you approve it. For current prices and fundamentals, a market-data server (Finnhub, Polygon.io, Alpha Vantage, or Financial Modeling Prep). For filings and sourced analysis, a research server. Much of the field is open-source and self-hostable. Pick by whether you want account access, live data, or sourced research, and expect to run more than one. Walnut is not an investment adviser.