mcp-native · quality-scored · real production code

Give your agent
134,011 real implementations

Semantic search over quality-scored components mined from real production repos — not a synthetic style guide. Connect Claude Code, Cursor, or Codex directly.

134,011

components indexed

96,567

with full source

Why it's worth connecting

real production provenance

Every component is mined from an actual public repo, not generated on the fly — you get code that already works somewhere, with a GitHub permalink back to its source.

quality-scored, not guesswork

Each component carries 5 sub-scores (completeness, code quality, deployability, reusability, strategic value) rolled into one composite — filter by min_composite and only see the good stuff.

no context-switching

MCP-native means your agent searches this library from inside Claude Code, Cursor, or Codex directly — no separate browser tab, no copy-paste round trip.

scale a hand-curated kit can't match

40K+ real components across 16 categories — bigger and more varied than any single component library, because it's pulled from thousands of real codebases, not one.

How it works

step 1

Get a free API key below — no credit card. 15 free credits to try it, then 5 free searches every day (resets daily, unused credits don't carry over).

step 2

Add it to your MCP client config (Claude Code, Cursor, Codex, etc):

{
  "mcpServers": {
    "cobalt-components": {
      "url": "https://15-204-92-178.nip.io/mcp",
      "headers": { "Authorization": "Bearer YOUR_API_KEY" }
    }
  }
}

step 3

Your agent can now call search_components, get_component, and library_stats directly — no copy-pasting code from a browser tab.

Free tier: 15 credits to try it, then 5 free searches every day via MCP — resets daily, unused credits don't carry over. No credit card.

Get a free API key