MCP live

awesome-repositories.com MCP server.

Point any MCP-aware agent at https://awesome-repositories.com/api/mcp. The first tool call walks you through a one-click sign-in with Google; after that your agent can search the directory by intent, read any repository, and find its documentation. It's a read-only server — search and read, nothing to write — and every tool is deterministic: the retrieval is ours, the judgement stays with your agent.

Recommended setup

Step 1Install the research skill

Start here. One command installs the awesome-repositories skill into every agent you have, resolved through the well-known index at /.well-known/agent-skills/index.json. The skill carries the deep-research playbook this site runs on: work out the intent, resolve it to canonical taxonomy tags, retrieve from several arms, verify every result against its stored tag evidence, and ask a clarifying question when the results are ambiguous. Step 2 connects the server the skill drives.

npx skills add https://awesome-repositories.com

Step 2Connect the server — one command, every agent

Adds the server to every MCP client it detects — Claude Desktop, Claude Code, Cursor, VS Code, Windsurf, Cline, Codex, Zed, Goose, Copilot CLI, and more. Append -y to skip prompts, -g for a global (user-level) install, or --agent <client> to target one.

npx add-mcp https://awesome-repositories.com/api/mcp --name awesome-repositories

Or add it per client

Claude Code

claude mcp add --transport http awesome-repositories https://awesome-repositories.com/api/mcp

VS Code

code --add-mcp '{"name":"awesome-repositories","type":"http","url":"https://awesome-repositories.com/api/mcp"}'

Cursor / Codex / Windsurf / Zed

npx add-mcp https://awesome-repositories.com/api/mcp --agent cursor   # or: codex · windsurf · zed · cline · goose

Claude Desktop / any other client — paste into its MCP config

{
  "mcpServers": {
    "awesome-repositories": {
      "type": "http",
      "url": "https://awesome-repositories.com/api/mcp"
    }
  }
}

No Authorization header — the client handles the OAuth dance for you.

Authentication

OAuth 2.1 — there's no token to paste. The client auto-discovers the authorization server from /.well-known/oauth-protected-resource, registers itself (Dynamic Client Registration), then opens a browser tab so you can sign in with Google. Refresh tokens keep the session alive. Every signed-in user can call every tool — this is a read-only server, so there's nothing to gate by role.

Tools

All eight are read-only and deterministic — no AI re-ranking, no quota. Available to any signed-in user.

search_repositories — deterministic hybrid retrieval — BM25 over repositories fused with vector search (query→tag→repo and query→repo-centroid) by Reciprocal Rank Fusion. Narrow with filter_tag_paths; the clusters summary shows which taxonomy roots the results landed in.
find_repository — resolve a repository you already know by name, owner/repo, or GitHub URL — description, tagline, language, topics, and its canonical tag paths, plus near name matches.
search_metadata_tags — hybrid BM25 + vector search over the unified tag tree — turn a concept into a canonical tag path, with a tree neighbourhood around every hit.
explore_tag — everything around a known tag path — ancestors, siblings, children, subtree repo count, co-occurring tags, and the top repositories under it.
match_curated_searches — match an intent against the editor-curated search pages and their vetted related tags, each scored 0–1 by confidence.
get_repository — full detail for one repository by internal id or owner/repo, including every public tag assignment with its correlation score, why-it-matches note, and cited source excerpts.
list_curated_searches — admin-pinned search slugs — handy discovery starting points when there is no query yet.
list_repository_docs — the documentation pages indexed for a repository (readme / doc), each with its source URL — fetch a page's body directly from that URL.