May 29, 2026

A Free MCP Server That Gives AI Agents 12 Live Financial Data Tools

Why we built one MCP endpoint for SEC insider trading, 13F, 8-K, FDA, congress trading and crypto data, how the proxy-over-standby architecture works, and the gotchas.

Most AI agents are great at reasoning and bad at knowing things. Ask one "which companies had insider buying clusters this week?" and it will either hallucinate or try to scrape the web and time out.

The boring fix: an MCP (Model Context Protocol) server that exposes 12 scored financial and market-data tools behind one endpoint, so Claude, Cursor or any MCP-capable agent can pull real, structured, source-linked data on demand.

What the agent gets

One connection, twelve tools: insider buying clusters (SEC Form 4), planned insider sales (Form 144), hedge-fund 13F consensus, 8-K material events, 13D/G activist stakes (including the activist's stated purpose from Item 4), biotech trial catalysts, FDA drug actions, US government contract awards, congress trading, NIH research funding, a crypto momentum scanner and App Store review intelligence.

All sources are official and public - SEC EDGAR, openFDA, ClinicalTrials.gov, USAspending.gov, House Clerk, NIH RePORTER, CoinGecko, Apple App Store. Every result links back to the original filing, so the agent can cite its source instead of asserting from memory.

Why MCP instead of a REST API

Integration cost. With a REST API, every data source means auth, schema reading and glue code. With MCP, the agent discovers the tools and reads their descriptions - the descriptions ARE the documentation. We write them for the model ("use this when the user asks about activist investors..."), not for humans.

The server itself is free. The underlying data tools bill $0.20 per result through Apify's pay-per-event model, only when a tool is actually called. No subscription, which matters for agents that might query a feed twice a month. Charging at both the server layer and the data layer would punish exactly the users we want.

The architecture: a proxy over standby

The server runs as an Apify Actor in standby mode - a persistent process speaking streamable HTTP MCP at /mcp. Each tool is a thin proxy that calls the corresponding data Actor and returns its dataset. We chose proxying over reimplementing the logic in the server for three reasons: one source of truth (the data Actors already run and are monitored daily), independent billing per tool, and crash isolation - a bug in one parser cannot take down the endpoint.

The underlying data tools run as transparent Apify Actors, each monitored daily with a source link on every result.

Gotchas that cost real time

  • FastMCP 3.x positional arguments: FastMCP("name", "x") silently treats the second positional as instructions. Use keyword arguments.
  • Standby permissions: Apify standby Actors run with limited permissions but MAY call other Actors - exactly what a proxy needs, not obvious from the docs.
  • Memory limits: builds share an 8 GB pool on the default plan; push Actors sequentially or builds block each other.

Try it

The endpoint is https://datasignalslab--datasignals-mcp.apify.actor/mcp - add it to Claude Desktop, Claude Code or Cursor as a remote MCP server with an Apify token, and ask something like: "Check insider buying clusters from the past week, and for the top company also check recent 8-K events and congress trades." The agent chains three tools and answers with scored, source-linked data.

All tools are documented on the MCP server page and the Apify listing.

Financial tools return data for research, screening and monitoring - not investment advice.

Prefer the tool to the write-up? Browse all data feeds or connect the free MCP server.