mcp server

mcprated

Open quality ratings for Model Context Protocol servers — daily lint, 4 axes (Reliability/Documentation/Trust/Community), composite 0-100, MIT ruleset.

mcprated/mcpratedcommunitylistedunreviewed

Use tool

Set up

Runs on your computer. Pick your agent and operating system.

This project does not publish an install method we can turn into a config. Follow the README instead.

README

synced Oct 21, 2026

MCPRated

Agent-readable index of every MCP server. Built for LLMs to discover, vet, and choose tools at runtime — not for humans to browse.

Open ruleset, daily-updated, deterministic. Static catalog + remote MCP endpoint, both live.

🌐 Live

MCP server: https://mcp.mcprated.workers.dev (streamable HTTP)
Static catalog: mcprated.github.io/mcprated

Add to your client in one line

Claude Code / Claude Desktop:

claude mcp add --transport http mcprated https://mcp.mcprated.workers.dev

Cursor / Cline / Continue / any client supporting remote MCP: paste into your config:

{"mcpServers": {"mcprated": {"url": "https://mcp.mcprated.workers.dev"}}}

The agent then has access to 8 tools:

Tool	When to call
`find_server`	Map your need to a capability category (12 of them) and rank servers by quality
`search`	Free-text search when the capability enum doesn't fit
`find_tool`	Discover a specific tool by name (`browser_navigate`, `read_file`, …) across all servers
`vet`	Trust verdict for one server (verified / caution / low_quality) before installing
`alternatives`	Capability-similar fallbacks when a server is unavailable
`by_kind`	Filter classifier output (server / client / framework / tool / ambiguous)
`top`	Top servers by composite, stars, or recency
`server_detail`	Full lint signal breakdown — every signal, every reason

See https://mcp.mcprated.workers.dev for a live tool list, or /api/v1/manifest.json for the JSON schema.

Static catalog API

Fallback for agents without remote-MCP support, or for direct integration without an MCP client. No auth, stable URLs, daily refresh.

curl https://mcprated.github.io/mcprated/llms.txt                            # discovery + endpoint map
curl https://mcprated.github.io/mcprated/api/v1/manifest.json                # full endpoint schema
curl https://mcprated.github.io/mcprated/api/v1/by-capability/database.json  # one shard per category
curl https://mcprated.github.io/mcprated/api/v1/vet/<owner>__<repo>.json     # trust subset
curl https://mcprated.github.io/mcprated/api/v1/tools-index.json             # flat list of every extracted tool
curl https://mcprated.github.io/mcprated/index.json                          # full catalog
curl https://mcprated.github.io/mcprated/servers/<owner>__<repo>.json        # per-server detail
curl https://mcprated.github.io/mcprated/excluded.json                       # transparency

Per server we publish: composite score, four axis scores, kind (server / client / framework / tool / ambiguous), subkind (integration / aggregator / prompt-tool / agent-product), capabilities[], distribution, tool_count, tool_names_preview, hard flags, license, language, recency.

Daily snapshots in Releases — historical state retained.

Run locally (for development)

# In one terminal — local Worker
cd worker
npm install
npm run dev      # starts MCP server on http://localhost:8787

# In another terminal — connect a client
claude mcp add --transport http mcprated-local http://localhost:8787

# Or test interactively with the official MCP Inspector
npx @modelcontextprotocol/inspector@latest http://localhost:8787

The local Worker uses the same source as production. Test changes locally; auto-deploy fires on push to main (gated by 260+ tests).

Smoke harness — see what your change did, before push

After every scoring/extractor change, run the local smoke harness. It lints every entry in tests/regression/seed.txt and prints a diff against the last snapshot — no judgment, just data.

python3 linter/smoke.py            # first run — snapshot only
# ... make changes to linter/ ...
python3 linter/smoke.py            # second run — diff vs last snapshot

Output: a current-state table (composite, kind, capabilities, tool_count, hard_flags per server) and a diff section listing only what changed. Last 10 snapshots are kept under .local/smoke/.

Use cases:

Verify a taxonomy keyword change didn't shift unrelated capabilities
Confirm an extractor patch raises tool_count for the targeted servers
Catch trust-axis regressions when adding/removing signals
Check the Tier B junk repos still score low

The script reuses the shared .cache/ so it's fast (~30s after warm cache). Add GITHUB_TOKEN=$(gh auth token) to fetch missing entries authenticated.

What we catalog

An MCP server is a runnable artifact implementing the Model Context Protocol (stdio / SSE / streamable HTTP) that exposes ≥1 of tools / resources / prompts and is distributed as a product for use by an MCP client. We do not catalog frameworks for building MCP servers (FastMCP, official SDKs), MCP clients/inspectors, or end-user apps that consume MCP without exposing it.

See methodology.md for the full operational definition and linter/taxonomy/v1.yaml for the capability vocabulary.

Score model: 4 axes × 0–100 → composite 0–100. Hard flags can cap composite (archived → 30, empty_description → 75, …).

Axis	Question
Reliability	Will it work and keep working?
Documentation	Can a stranger figure this out?
Trust	Safe to depend on?
Community	Are people caring for it?

Open ruleset (MIT) at linter/rules/v1.0/. Open data (CC-BY-4.0). Versioned: every score carries rule_set_version.

For maintainers — embed your badge

[![MCPRated](https://mcprated.github.io/mcprated/badges/<owner>__<repo>.svg)](https://mcprated.github.io/mcprated/)

Replace <owner>__<repo> with double-underscore-encoded path. Example:

[![MCPRated](https://mcprated.github.io/mcprated/badges/microsoft__playwright-mcp.svg)](https://mcprated.github.io/mcprated/)

Renders as: MCPRated

Badge always reflects the latest score. No URL pinning, no stale embeds.

Status

v0.2 — agent-first core. Reference seed ~25 servers + topic-search crawl ~125 indexed. Daily lint running. kind + capabilities shipped in rule_set v1.1. Coming next: agent-shaped endpoints (/api/v1/find, /vet, /by-capability/<cap>) and @mcprated/mcp-server as an installable npm package.

Run lint locally

git clone https://github.com/mcprated/mcprated && cd mcprated
export GITHUB_TOKEN=ghp_...                                # 5000/h authed
python linter/crawler.py --seed tests/regression/seed.txt
python linter/lint.py

Docs

methodology.md — score model, hard flags, what counts as an MCP server, versioning policy
linter/rules/v1.0/ — open YAML ruleset (4 files, 20 signals)
linter/taxonomy/v1.yaml — capability vocabulary (12 categories)
CHANGELOG.md — rule-set version history

License

MIT (code) · CC-BY-4.0 (data in data/ and release snapshots) · See LICENSE, DATA-LICENSE