# 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. We catalog runnable artifacts that implement Model Context Protocol; we do not catalog frameworks for building MCP servers, MCP clients/inspectors, or end-user apps that consume MCP without exposing it.

## Use as MCP server (recommended for agents)

Endpoint: `https://mcp.mcprated.workers.dev` (streamable HTTP)

Install in your MCP client:
```
# Claude Code / Claude Desktop
claude mcp add --transport http mcprated https://mcp.mcprated.workers.dev

# Cursor / Cline / Continue — paste into mcp config
{"mcpServers": {"mcprated": {"url": "https://mcp.mcprated.workers.dev"}}}
```

8 tools: `find_server`, `search`, `find_tool`, `vet`, `alternatives`, `by_kind`, `top`, `server_detail`. Live tool schemas at https://mcprated.github.io/mcprated/api/v1/manifest.json.

## Stats

- Servers rated: 32
- Rule set: v1.3.2
- Taxonomy: v1.0
- Last update: 2026-05-05T06:29:12+00:00

## API (static JSON, public, free, no auth)

**Start here:**
- [Manifest](https://mcprated.github.io/mcprated/api/v1/manifest.json): endpoint map, valid enum values (capabilities, kinds), MCP tool definitions. **Fetch this first to learn the shape.**

**Sharded answers (one question per file, agent-friendly):**
- `/api/v1/by-capability/<cap>.json` — servers tagged with capability X (database, web, search, ...)
- `/api/v1/by-kind/<kind>.json` — server / client / framework / tool / ambiguous
- `/api/v1/top.json` — top-by-composite, top-by-stars, top-by-recency (25 each)
- `/api/v1/vet/<owner>__<repo>.json` — trust-focused subset + verdict (verified / caution / low_quality)
- `/api/v1/alternatives/<owner>__<repo>.json` — capability-similar servers, ranked

**Full views:**
- [Catalog index](https://mcprated.github.io/mcprated/index.json): every server with composite, axes, `kind`, `subkind`, `capabilities`, `distribution`
- [Per-server detail](https://mcprated.github.io/mcprated/servers/microsoft__playwright-mcp.json): full lint output for one repo (replace path with `<owner>__<repo>.json`)
- [Excluded list](https://mcprated.github.io/mcprated/excluded.json): repos filtered out by prefilter, with reason — transparency
- [Daily snapshot archive](https://github.com/mcprated/mcprated/releases): tarballs of historical state, retained forever

## Per-server fields agents care about

- `kind`: server | client | framework | tool | ambiguous
- `subkind` (when `kind=server`): integration | aggregator | prompt-tool | agent-product
- `capabilities[]`: top tags from taxonomy v1.0: database, filesystem, web, search, productivity, comms, devtools, cloud, ai, memory, finance, media (or empty if no match)
- `distribution`: how the artifact reaches a client (`repo` for v1; `npm`/`pypi`/`docker`/`hosted` reserved for upcoming ingest)
- `composite` 0–100 + 4 axis scores; `hard_flags[]` for caps (archived → 30, etc.)

## Score model

- 4 axes, each 0–100: Reliability, Documentation, Trust, Community
- Composite = mean of axes
- Hard flags cap composite (archived → 30, empty_description → 75, etc.)
- Color hints: 90+ green, 50–89 yellow, <50 red

## Coming next (planned)

- Agent-shaped endpoints: `/api/v1/find?capability=<cap>`, `/vet/<slug>`, `/by-capability/<cap>.json`, `/api/v1/manifest.json`
- `@mcprated/mcp-server` (npm) — installable MCP server that queries the catalog from your client

## Source

- Repo: https://github.com/mcprated/mcprated
- Methodology + operational definition of "MCP server": https://github.com/mcprated/mcprated/blob/main/methodology.md
- Ruleset (YAML, MIT): https://github.com/mcprated/mcprated/tree/main/linter/rules/v1.0
- Capability taxonomy: https://github.com/mcprated/mcprated/blob/main/linter/taxonomy/v1.yaml
- Changelog: https://github.com/mcprated/mcprated/blob/main/CHANGELOG.md

## License

- Code: MIT
- Data: CC-BY-4.0 (attribution: "Quality data from MCPRated, https://mcprated.github.io/mcprated/")