Open-source MCP server · v0.8.2
~/jobsync zsh
❯ cat ./README.md

Your MCP client is the career agent. We're the plumbing.

jobsync-mcp exposes deterministic tools — filters, dedup, ATS fetchers, fit-scoring — to any MCP client. Drive it from Claude Code, Claude Desktop, or Cursor.

npm i -g jobsync-mcp 1,400+ downloads MIT
npm install → Star on GitHub ↗
See it in action

One prompt. Fresh jobs. Scored against your resume.

Run /mcp__jobsync__scrape_jobs_workflow inside your MCP client. The LLM composes queries, fetches postings, and calls jobsync-mcp's deterministic tools to classify, dedup, verify, upsert, and fit-score — end to end.

Step 1 - Initialising
live
Fit analysis results produced by the agent
Step 1 Initialising
0:00
jobsync-mcp — scrape_jobs_workflow
v0.8.2
❯ /mcp__jobsync__scrape_jobs_workflow ◆ web_search ── composing queries from portals.yml 12 sources ◆ web_fetch ── extracting postings 87 candidates │ ↳ classify_job_batch · verify_job_link_batch · cache_is_seen ◆ airtable_upsert_job ── writing authoritative sink ◆ analyze_job_fit ── scoring against skills.md · experience.md ✓ synced 14 new roles to Airtable · avg fit 7.6 / 10
17 tools exposed
MCP surface
4 ATS fast-paths
Greenhouse · Lever · Ashby · Workday
SQLite dedup cache
~/.jobsync/cache.db
Architecture

Model-agnostic by design.

The MCP client's own model does the semantic work — composing searches, extracting postings, producing fit explanations. jobsync-mcp never ships an LLM of its own.

Everything deterministic — filters, SQLite dedup, link verification, ATS pulls, profile I/O — lives behind the MCP surface. Boring, testable, portable.

01 · your MCP client
Your MCP client
Claude Code · Desktop · Cursor
  • web_search
  • web_fetch
  • reasoning
  • fit analysis
02 · deterministic plumbing
jobsync-mcp tools
deterministic plumbing
  • filter · classify · dedup
  • verify_job_link
  • ats fast-paths
  • pipeline_upsert_jobs
03 · authoritative sink
Airtable · Markdown
authoritative sink
  • synced roles
  • fit scores 1–10
  • pipeline status
  • applied → offer
Quick start

Three commands. You're in.

01
Install
$ npm i -g jobsync-mcp
02
Configure
$ jobsync-mcp init
03
Parse resume
$ jobsync-mcp onboard --resume ~/cv.pdf
Register with Claude Code:
$ claude mcp add --scope user jobsync jobsync-mcp
Full setup guide →
Who built this

Built by Mayank Vyas, an ML Research Engineer at the CoRAL Lab at Arizona State University — advised by Prof. Vivek Gupta — during his own job search.

The deterministic filter logic and extraction schema are ported from a production pipeline. The agentic workflow is what replaced it. Open-sourced because every job search needs the same plumbing, and nobody should write that pipeline twice.

GitHub profile ↗ CoRAL Lab ↗
Feedback

Share your thoughts.

Found a bug, have a feature request, or just want to say hi? Fill out the form below.