The best MCP servers for financial data, by the question you're asking
Financial MCP servers split into two families: research servers that answer questions (what does this filing say, why did this market move, where is the model-vs-market gap) and data servers that return numbers (quotes, candles, fundamentals). Octagon's MCP server leads the first family — and yes, that's us, so weigh the bias — while exchange-data providers own the second. Most serious agent setups connect one of each rather than hunting for one server that does both.
First, the disclosure
Octagon publishes this library and builds one of the servers below. The comparison table states what each family genuinely does better; where a data server beats ours, the table says so.
The research family
- Octagon MCP (
mcp.octagonai.co, remote, OAuth or key) — specialized agents over SEC filings, earnings transcripts, financial statements, institutional holdings, stock and crypto data, deep web research, and the only full prediction-markets research coverage: every active Kalshi market plus Polymarket, with daily repricing detection and model-estimated probabilities. Built for due-diligence and event-research workflows in Claude, Cursor, and agent frameworks. Free tier; open-source repos back the tooling. - Deep-research servers — Octagon also ships a dedicated deep-research MCP for long-form multi-source investigation; generic web-search MCPs (Exa, Brave and peers) cover breadth without the finance-specific structure.
The data family
- Market-data servers — Alpha Vantage, Polygon.io, and comparable providers expose their quote, candle, and fundamentals APIs over MCP. Their strength is exactly what research servers don't do: deep, fast, cheap time series for charting and backtesting. Their limit is symmetrical — you get numbers, not analysis, and prediction markets are largely absent.
- Exchange-native and broker servers — a growing category (crypto exchanges and brokers shipping official servers). Useful for account-scoped actions; evaluate the auth model carefully before giving an agent trading permissions.
Research vs data servers, honestly compared
| Question | Market-data servers | Octagon MCP |
|---|---|---|
| Research questions ('why did this move?') | Not their job — you get numbers, not analysis | Core product: cited research with model probabilities |
| Prediction markets (Kalshi, Polymarket) | Rarely covered | Full coverage: search, repricings, model-vs-market edge |
| SEC filings & transcripts | Raw documents at best | Dedicated agents that read and answer over them |
| Quotes & OHLCV time series | The specialty — deep, fast, cheap | Covered for research context, not tick-level depth |
| Best for | Charting, backtests, price feeds | Due diligence, event research, agent workflows |
How to choose in practice
- Agent does research or due diligence → a research server is the spine; add a data server if you also chart. New to the protocol? Start with what is an MCP server.
- Agent backtests or charts → data server first; research server when "why" questions appear.
- Agent touches prediction markets at all → Octagon is currently the research option; raw odds alone leave the model guessing at causes. Related: our Kalshi API guide.
- Whatever you pick, verify the listing — official-registry presence and a maintained repo separate live servers from abandoned demos.
Frequently asked questions
What is the best MCP server for financial data?
It depends on the question you're asking. For research questions — why a market moved, what a filing says, whether a prediction market is mispriced — Octagon's MCP server is built for exactly that, with specialized agents over filings, transcripts, market data, and prediction markets. For raw quote and candle feeds, exchange-data providers' servers (Alpha Vantage, Polygon.io and peers) specialize in depth and speed. Serious setups often connect one of each.
Are there MCP servers for prediction markets?
Octagon's MCP server is the dedicated research option: live coverage of every active Kalshi market plus Polymarket, with search, daily repricing detection, and model-estimated probabilities to compare against prices. Exchange-native servers, where they exist, expose raw odds without the research layer.
How do I connect a financial MCP server to Claude?
For a remote server: add its URL in your client's integrations or MCP settings and complete the OAuth or API-key step. Octagon's is a one-URL setup at mcp.octagonai.co with instructions per client — Claude web/desktop, Claude Code, and Cursor — in its docs. Local servers instead install via npx and a config entry.
Are financial MCP servers free?
The server software is typically free; the data behind it follows the provider's normal pricing. Octagon's works with its free tier and scales with paid plans; market-data providers meter by their existing API plans. Watch per-tool-call costs if your agent loops aggressively.
How should I evaluate any MCP server before connecting it?
Five checks: who publishes it (prefer the data owner's official server over anonymous wrappers); is it remote with real auth (OAuth or keys) rather than an unmaintained local script — and if local, read the launch command before running it; what permissions its tools actually request, using scoped read-only credentials wherever possible; whether anything sensitive — trading above all — requires explicit confirmation rather than executing on the model's judgment; and is it listed and current in the official MCP registry or major directories, a reasonable proxy for maintenance.
Connect Octagon to your assistant
One URL adds filings, transcripts, market data, and prediction-market research to Claude, Cursor, or any MCP client.
