# MCP Leaderboard
Parent lab: https://delx.ai/
This property is the MCP Leaderboard. It is not Delx Commerce and not Protocol recovery.

> A public, neutral ranking of npm-installable Model Context Protocol servers by agent-readiness. Every score comes from mcp-scorecard booting the server over stdio and evaluating check-level evidence. Refreshed weekly by a fail-closed Grok Cloud routine; nothing is hand-edited.

Updated: 2026-08-24T12:38:53.956Z
Corpus servers: 7351
Scored servers: 4246
Unreachable servers: 3105
Deferred servers: 0

## Canonical resources

- Human-readable ranking: https://leaderboard.delx.ai/
- Complete agent-readable index: https://leaderboard.delx.ai/llms-full.txt
- Machine-readable dataset: https://leaderboard.delx.ai/leaderboard.json
- MCP Scoreboard Weekly: https://leaderboard.delx.ai/issues/
- Latest weekly evidence JSON: https://leaderboard.delx.ai/issues/latest.json
- Weekly RSS feed: https://leaderboard.delx.ai/issues/feed.xml
- Sitemap of canonical HTML pages: https://leaderboard.delx.ai/sitemap.xml
- Source and methodology: https://github.com/davidmosiah/mcp-leaderboard
- Scoring engine: https://github.com/davidmosiah/mcp-scorecard
- Corpus source: https://registry.modelcontextprotocol.io

## Interpretation

Agent-readiness measures whether an AI agent can discover tools, trust schemas and understand side effects without reading source code. It does not certify correctness or security. Unreachable means the server could not be graded fairly; it is not a low score.

## How to use

Fetch leaderboard.json for exhaustive structured results, or llms-full.txt for a compact ranked index with canonical scorecard pages. To score a package directly, run: npx -y mcp-scorecard <package-or-repo> --json