What it reads is one README in one repository, parsed into entries with name, description, author and category — so it answers what is listed, not what is currently running or maintained. That is still the fastest way to answer does something already exist for X before writing your own, and the feature search is better suited to that question than scrolling a long Markdown page.
An MCP server that queries the official `modelcontextprotocol/servers` repository, parsing its README into a searchable catalogue split across reference, official and community entries — 547 servers by that page's own headline count, which is only ever as fresh as its last edit.
- `list_servers` — browse the catalogue with an optional `category` of reference, official, community or all, a text `search`, and a `limit` defaulting to 20
- `get_server_details` — everything known about one server, by name
- `search_servers_by_feature` — find servers offering a capability such as database, api or blockchain, with a `limit` defaulting to 10
- `get_random_servers` — a random selection for browsing, `count` defaulting to 5
- `refresh_server_list` — force a refresh of the cached catalogue from GitHub
Nothing — no account, no key. The package is `@atonomus/mcp-servers-search` (1.0.1 in package.json), run with `npx -y @atonomus/mcp-servers-search` or installed globally. The catalogue is cached for 1 hour to keep GitHub API calls down; refresh manually when you need it fresher. MIT licensed.
One command — npm install -g @atonomus/mcp-servers-search