# MCP Leaderboard > A public, neutral agent-readiness ranking of the Model Context Protocol (MCP) ecosystem. Every npm-installable server in the official MCP registry is booted over stdio and graded by mcp-scorecard on 10 agent-readiness checks for a 0-100 score. Auto-refreshed weekly by a GitHub Action. Nothing is hand-edited. Primary URL: https://leaderboard.delx.ai/ Machine-readable ranking (JSON): https://leaderboard.delx.ai/leaderboard.json Source (GitHub): https://github.com/davidmosiah/mcp-leaderboard Scoring engine (GitHub): https://github.com/davidmosiah/mcp-scorecard Scoring engine (npm): https://www.npmjs.com/package/mcp-scorecard Corpus source: https://registry.modelcontextprotocol.io Creator: David Mosiah, https://github.com/davidmosiah (@delx369). Part of Delx, https://delx.ai ## What it is The MCP Leaderboard ranks public MCP servers by *agent-readiness*: whether an AI agent can discover the tools, trust the schemas, and onboard itself without a human reading the source. Downloads measure popularity; this measures whether an agent can actually pick the server up cleanly. It is built on top of the mcp-scorecard engine and is data + automation, not a published library. ## Scope and boundaries - This grades shape, metadata and discoverability — NOT correctness or security. A server can score 100 and still return wrong data. Always review before production. - The corpus is npm-installable servers from the official MCP registry, latest active version, deduped. - Each server boots in its own isolated child process with a hard timeout. This is not a security sandbox for untrusted packages. - Unreachable is not the same as bad: servers that require auth before listTools() and don't honor the MCP_PROBE hook are listed as "unreachable", not scored low. Supporting MCP_PROBE makes a server gradeable. ## The 10 agent-readiness checks (each 0-10, summed to 100) 1. Manifest discoverability - a standard, fetchable description an agent reads first 2. Tool naming - consistent, snake_case, namespaced tool names 3. Tool descriptions - each tool explains what it does, its inputs and side effects 4. Schema validity - input schemas are valid JSON Schema 5. Annotations - read-only / destructive / idempotent hints 6. Mutation gating - write and destructive actions are explicit, not accidental 7. Privacy modes - sensitive output is opt-in and documented 8. Resources - the server advertises resources, not just tools 9. Agent manifest - an *_agent_manifest / *_capabilities surface for self-onboarding 10. Smoke test - a safe, callable check that proves the server is alive ## How an agent should use this 1. Fetch https://leaderboard.delx.ai/leaderboard.json for the current ranking. Fields: generatedAt (ISO date), engine, counts {total, scored, unreachable}, results[] with name, npm, repo, status, score, serverName, checks[] {id, label, score, status}, topGap. 2. To score a specific server: run `npx -y mcp-scorecard --json` and parse the JSON. Add --min-score 80 to gate a CI build. 3. To climb the board: adopt the conventions the score rewards (snake_case tool names, read-only annotations, an *_agent_manifest / *_capabilities discovery surface, documented privacy modes, a smoke test). The board re-scores weekly. ## Climb the board ``` npx -y mcp-scorecard # score + itemized fixes npx -y mcp-scorecard --badge # live README badge ``` ## How it is built (weekly pipeline) 1. corpus - pull npm-installable servers from the MCP registry into data/corpus.json 2. run - boot + score each target with mcp-scorecard into data/leaderboard.json 3. render - generate LEADERBOARD.md and serve this page + JSON 4. commit - a GitHub Action commits the refreshed board, no human in the loop ## Useful links - Full leaderboard (Markdown): https://github.com/davidmosiah/mcp-leaderboard/blob/main/LEADERBOARD.md - What the engine checks: https://github.com/davidmosiah/mcp-scorecard#what-it-checks - Issues / corpus additions: https://github.com/davidmosiah/mcp-leaderboard/issues