Talk to JobMatch's job data directly.

Overview

MCP (Model Context Protocol) is an open standard for connecting an AI agent — Claude, or anything else that speaks it — to an external tool or data source in a structured way, instead of that agent having to scrape or guess at a website's HTML.

This server exposes three read-only tools over JobMatch's own listing and market-stats data: search by keyword, country, remote type, or tag; look up a single listing; or pull the aggregate market snapshot shown on the Job Analytics page. It reads only what's already public — nothing here is behind a login, and nothing exposed here isn't already visible on the site itself.

Quickstart

The server speaks the MCP Streamable HTTP transport. Point any MCP-compatible client at the endpoint above, or call it directly as JSON-RPC over HTTPS.

Option A — add it to an MCP client

For Claude Desktop, Claude Code, or any client that supports a remote MCP server by URL, add:

{
  "mcpServers": {
    "jobmatch": {
      "url": "https://jobmatch-mcp.kennethj.workers.dev/mcp"
    }
  }
}

Option B — call it directly

Any HTTP client works. A tool call is a single JSON-RPC request:

# search_jobs
curl https://jobmatch-mcp.kennethj.workers.dev/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "tools/call",
    "params": {
      "name": "search_jobs",
      "arguments": { "country": "Sweden", "tags": ["Manager"], "limit": 5 }
    }
  }'

No initialize handshake to manage by hand — the server is stateless, so every request is self-contained. An MCP client library will still perform the standard handshake automatically; that's expected and handled.

Authentication & rate limits

No API key, no login. Every request is anonymous and counted by IP address: 60 requests per hour, fixed window. Going over it returns an HTTP 429 with a plain JSON body:

{ "error": "Too many requests. Try again later." }

An opt-in API key for higher limits is planned but not built yet — see Roadmap.

Tools

All three tools are read-only and side-effect-free. Every job record's applyUrl routes through JobMatch's own click-tracked redirect rather than a raw scraped link — that's deliberate, and it's the same link the site itself renders.

toolsearch_jobs

Search active listings. All parameters are optional and combine with AND. Results are sorted newest-posted-first — this version does plain keyword matching, not semantic ranking (see Roadmap).

{
  "count": 143,
  "returned": 1,
  "results": [
    {
      "id": 2268,
      "jobTitle": "Operations Manager",
      "company": "Flow Intimates",
      "location": "Frederiksberg",
      "country": "Denmark",
      "city": "Frederiksberg",
      "remoteType": null,
      "employmentType": null,
      "salaryMin": null,
      "salaryMax": null,
      "currency": null,
      "sourceSite": "jobindex.dk",
      "datePosted": "2026-09-24",
      "tags": ["Manager"],
      "applyUrl": "https://jobmatch-worker.kennethj.workers.dev/go/2268"
    }
  ]
}

toolget_job

Look up a single listing by id — the same shape search_jobs returns per result.

An unknown id doesn't error — it returns isError: true with a plain message, so a bad id is a normal tool result to handle, not an exception.

toolget_market_stats

No parameters. Returns the same aggregate, anonymised snapshot shown on Job Analytics — totals, country/tag/seniority breakdowns, and week-over-week trend. No per-listing detail; use search_jobs for that.

Data & caching

• Source data refreshes on JobMatch's own scraping schedule — roughly every 2 days.
• Responses are cached at the edge for 5 minutes, so a burst of calls doesn't cause redundant fetches. This doesn't affect freshness beyond that window.
• Embeddings used for the site's own semantic search are never included in tool responses — they're meaningless outside that specific model and would only inflate every response.
• applyUrl is always a tracked redirect, never the raw scraped URL — clicking through counts the same way a click from the site itself does.

Terms

Free to use. If you build something public on top of this data, an attribution link back to kennethjensen.me/jobmatch is appreciated, not required. Don't use it to rebuild a competing scrape of the underlying sites JobMatch itself tracks — the point of this server is to be a better way to use JobMatch's own data, not a shortcut around everyone else's.

Roadmap

This is a phase-1 build. Not yet done, roughly in order:

• An opt-in free API key for higher rate limits.
• Semantic/embedding-ranked search, matching what the site's own search page does.
• A "new listings" subscription or webhook tool, so a real-time integration doesn't need to poll.

Something missing, or a rough edge you hit integrating? Get in touch.