Stop letting Claude hallucinate short interest.
AlgoThesis hands your AI the market data it can't see — FINRA short interest, grounded web research with ticker mapping, and a catalyst calendar. One install, every MCP-aware agent on your machine gets it.
Free · no API key · MIT on npm
Install
Claude Desktop
Edit ~/Library/Application Support/Claude/claude_desktop_config.json (Mac) or %APPDATA%\Claude\claude_desktop_config.json (Windows). Add this block, then quit + relaunch:
{
"mcpServers": {
"algothesis": {
"command": "npx",
"args": ["algothesis-mcp"]
}
}
}jsonClaude Code (CLI)
One command:
claude mcp add algothesis npx algothesis-mcpbash
Cursor / other MCP clients
In Cursor: Settings → MCP → Add Server → paste the same JSON block from step 1.
Any MCP-compatible client works — same config format.
What you get
research_topicResearch any topicLive Google Search + ticker mapping. Ask about people, events, macro themes — get a grounded summary with source URLs and tradeable tickers.
Example: "What's the latest on Kevin Warsh and how should I be positioned for a dovish Fed?"
get_short_interestReal short-interest dataFINRA short-interest data via Yahoo (~14-day lag) for up to 20 tickers in one call. Sorted by short % of float. The data ChatGPT doesn't have.
Example: "Show me real short interest on CVNA, BYND, GME, AMC — which has actual squeeze potential?"
find_catalystsThe catalyst calendar your AI doesn't haveEarnings, FDA approvals, Fed meetings, market-moving events — by ticker or sector. The dates Claude otherwise has to guess at.
Example: "Which biotechs have FDA catalysts this month?"
Real prompts that work today
Paste any of these into Claude after install. The model picks the tool.
Why this exists
LLMs invent short-interest numbers. They miss catalysts that haven't been blogged yet. Tool calls fix both — the model stops guessing and starts calling FINRA. This is the same data the AlgoThesis web app runs on; the MCP server hands it to whichever agent you already use.
Source on npm · MIT · contributions welcome.