ToolCairnThe problem it solves
Plain web search and an LLM's training data are a poor combination for tool selection. Knowledge cutoffs miss recent releases, search engines surface tutorials over authoritative ranking, and version-compatibility answers live in scattered issue threads. Agents end up guessing — and the user pays for the guess in a broken upgrade two weeks later.
ToolCairn closes that gap with a continuously-updated index of 30,000+ open-source tools across 35+ registries (npm, PyPI, Cargo, Maven, Go, Composer, RubyGems, NuGet, Homebrew, and more). The agent asks; ToolCairn answers with ranked candidates, alternatives, version-aware compatibility checks, and health signals it can quote.
Concretely, what does that look like?
You ask your agent: "I need a fast HTTP client for Node." The agent calls search_tools; ToolCairn returns ranked candidates with maintenance signals, alternatives, and a credibility flag if anything looks stale.
{
"ok": true,
"data": {
"matches": [
{ "name": "undici", "score": 0.91, "tier": "active" },
{ "name": "got", "score": 0.84, "tier": "active" },
{ "name": "ky", "score": 0.78, "tier": "active" }
],
"alternatives": ["axios", "node-fetch"]
}
}The agent weaves that into its reply — usually a short recommendation with the rationale ToolCairn surfaced. No more hand-wavy "axios is popular" blanket advice.
Plug it in
The MCP server is published as @neurynae/toolcairn-mcp on npm (currently v0.10.24). Most clients pick it up via a one-line config edit. For Claude Code:
claude mcp add toolcairn -- npx @neurynae/toolcairn-mcpRestart your agent. The first call opens a browser tab for sign-in; after that, every tool is available — no further setup. Other MCP clients (Cursor, Windsurf, VS Code Copilot, custom SDKs) follow the same pattern.