agent-called-whatWhat AI agents actually call

Behavior-based rankings of MCP servers, agent skills & libraries — not star counters.

📊 Behavior, not hype

We count how often each MCP server appears in real public mcp.json configs — not how many stars a README collected.

Every update tracks week-over-week movement — what's rising, what's fading, which packages are new. The ecosystem's pulse, not just a snapshot.

Every snapshot is a versioned JSON file in /data. Clone the repo, run three Python scripts, reproduce the rankings bit-for-bit.

Scripts use only Python stdlib + urllib. No pip install, no lockfile, no supply-chain anxiety.

Composite formula and every signal source are documented in methodology.md. Weights are tunable via PR.

New MCP server? Add one line to mcp_packages.py. Better weights? Open an issue. This is open source in the literal sense.