Summary
It scores your definitions, not your code.
This looks at the strings an agent actually sees — tool name, description, input schema — and tells you which of the four dimensions is dragging the score down. The project's own scan puts the problem in context: of 4,162 registered servers it looked at, 1,122 had tool definitions at all.
What it is
ToolRank scores a server's tool definitions from 0 to 100 across four weighted dimensions and places the result on a maturity scale. It ships as a browser scorer, a Python package and an MCP server.
What you get
- A 0-100 score across four dimensions: Findability at 25%, Clarity at 35%, Precision at 25% and Efficiency at 15%
- A maturity band rather than a bare number — Dominant at 85-100, Preferred at 70-84, Selectable at 50-69, Visible at 25-49, Absent at 0-24
- A formatted report you can act on: the project's worked example moves a definition from 52 to 96 by renaming the tool, describing what it returns, and typing its input schema
- The same score three ways — in the browser, from Python, or over MCP
Requirements
Run the MCP server with `npx @toolrank/mcp-server`; no account or key is described for it. The scoring engine is open source, and the browser scorer takes either your tool JSON pasted in or a Smithery server name.
Setup effort
One command — npx @toolrank/mcp-server