ResuRank MCP

A Model Context Protocol server that lets Claude Desktop score how well a resume matches a job posting, entirely on-device. It’s a companion to ResuRank, sharing the same @resurank/scoring engine, packaged as a local stdio MCP server instead of a desktop app.

Point it at a resume file once via RESUME_PATH, then ask Claude to score any job description against it. No API keys, no uploads — a ~25 MB embedding model downloads once into the Hugging Face cache, and every call after that runs locally in under a second.

Features

• One tool, resurank_score — scores a resume against a job title and description, returning a 0–100 match score with a full breakdown
• Hybrid scoring — 60% semantic embedding similarity + 40% keyword TF-IDF, with a divergence penalty that guards against false positives
• Fully local — resume text and job descriptions never leave the machine; the embedding model runs on-device via Transformers.js
• Live reload — edits to the resume file on disk are picked up on the next call, no restart required
• Inline overrides — pass a different resume_path or raw resume_text per call to test resume variants without touching RESUME_PATH
• One-command installer for macOS/Linux, or a manual npx setup for Windows
• Automatic setup from the ResuRank desktop app, which locates Node.js and writes the Claude Desktop config for you