recruiter-lens

Audits GitHub repos and profiles for hiring signal — as an MCP server (audit_repo, recruiter_scan tools) usable from Claude Code, Claude Desktop, or any MCP client, and as a standalone CLI.

Packages the repository-auditor and technical-recruiter-review skill methodology as a tool any AI assistant can call directly, instead of a prompt you paste in.

Design choice worth knowing about

This tool does not call an LLM itself. It gathers deterministic, checkable facts from the GitHub API (README presence, CI, license, topics, commit-message patterns, activity recency, bio/pin presence) and applies rule-based severity/verdict logic. The qualitative judgment — actually reading the code, weighing architecture — is left to whatever AI assistant is calling this tool.

This is a deliberate simplicity choice, not a missing feature: the task here is fact-gathering, and a second nested LLM call would add cost and non-determinism without adding real capability the calling assistant doesn't already have. It also means the entire tool is testable without an API key — see test/.

MCP usage

claude mcp add recruiter-lens -e GITHUB_TOKEN=<your token> -- npx recruiter-lens-mcp

Then ask your assistant something like "use recruiter_scan to check github.com/someuser" or "audit_repo on octocat/Hello-World".

CLI usage

GITHUB_TOKEN=<token> npx recruiter-lens audit-repo <owner>/<repo>
GITHUB_TOKEN=<token> npx recruiter-lens scan <username>

GITHUB_TOKEN needs at least public read access; a classic PAT with repo scope covers private repos too.

Testing

npm test runs the scoring-logic tests (test/*.test.mjs) — pure functions, no network, no API key required, runs in CI on every push.

License

MIT — see LICENSE.