Massive MCP: market data an agent can query with SQL

The usual problem with market analysis in a chat is volume: a year of daily bars or snapshots for a few hundred tickers fills the context fast. The official MCP server from Massive.com (formerly Polygon.io) takes a different route. Results can be stored as in-memory SQLite tables, the agent filters and aggregates them with SQL, and only the answer comes back into the conversation. The repo had about 400 stars as of 2026-10-06.

What it does

• Three tools for the whole API: search_endpoints finds endpoints in plain language, call_api calls any Massive REST endpoint, and query_data runs SQL (CTEs and window functions included) over stored tables. The endpoint index is read from Massive’s llms.txt at startup, so new APIs show up without a server release.
• Coverage: aggregates for stocks, options, forex, crypto and futures; real-time and historical trades and quotes; snapshots and top movers; dividends, splits and IPOs; fundamentals; Benzinga analyst ratings and news; Treasury yields and inflation data.
• Built-in finance functions: the apply parameter computes Black-Scholes prices and Greeks, simple and log returns, Sharpe and Sortino ratios, SMA and EMA.

Who it’s for

• Analysts writing research notes or weekly market recaps who want the agent to pull history and compute returns and volatility in one go.
• Hobbyist quants checking questions like “what was this sector’s Sharpe ratio over the past year” without writing a script.
• Options traders who want Greeks computed across a chain.

Setup

You need Python 3.12+, uv 0.4.0+ and a Massive.com API key. The README recommends installing it as a tool first to avoid cold-start timeouts, then registering it with Claude Code:

uv tool install "mcp_massive @ git+https://github.com/massive-com/mcp_massive@v0.10.0"
claude mcp add massive -e MASSIVE_API_KEY=your_api_key_here -- mcp_massive

Our take

Plenty of finance servers just wrap yfinance, with no service guarantee behind the data. Massive is a real market-data vendor, and the store-then-query design is the clearest answer we’ve seen to the too-much-data problem. Three caveats: the project calls itself experimental and may break between versions, what you can fetch depends on your Massive account, and it only reads and computes. It places no orders and offers no investment advice. The old POLYGON_API_KEY variable still works. MIT-licensed.