Job Search MCP

What Servation/job-search-mcp MCP server does

Servation/job-search-mcp MCP server combines job discovery, fit evaluation, and application triage in a local workflow. Searches cover LinkedIn and eight additional sources: Greenhouse, Lever, Ashby, Workday, SmartRecruiters, Hacker News, RemoteOK, and Remotive. Queries can target roles and locations, with optional filters for seniority, employment type, posting age, remote status, salary, applicant count, and source.

After a search, Claude reads the available job descriptions and assigns each listing a fit score from 0 to 100, along with a short reason. The resulting shortlist can be displayed as an inline board. Listings can be applied to, skipped, or saved individually, and multiple entries can be triaged by score or source. The project is tuned for US software-engineering roles, so searches outside that scope may return limited results.

How it works

The MCP server runs locally over stdio and communicates with Claude Desktop. Its accompanying MCP App UI is bundled as a single HTML resource that renders the ranked board inside a compatible host. Claude selects the exposed tools during normal conversation; users can request actions such as finding jobs, showing the board, or reviewing saved applications without manually invoking tool names.

A profile can be saved from a resume. The profile supplies skills, roles, experience, and location information used by later searches and scoring. The host model performs the normal evaluation directly, so the project does not require a separate LLM API key. A deterministic scoring fallback exists in the source for weaker local models, but it is not enabled by default.

The board acts as a persistent shortlist. New searches avoid jobs already on the board, in the tracker, dismissed, or shown within the previous six months. Applied jobs move to the tracker, while skipped jobs are hidden. The data is stored in one JSON file on the local machine and is not sent to a service other than the job boards being searched.

Setup and configuration

Servation/job-search-mcp MCP server can be added to Claude Desktop with the published npm package:

{
  "mcpServers": {
    "job-search": {
      "command": "npx",
      "args": ["-y", "@servation/job-search-mcp"]
    }
  }
}

After changing the configuration, fully quit and reopen Claude Desktop. The documented configuration locations are %APPDATA%\\Claude\\claude_desktop_config.json on Windows and ~/Library/Application Support/Claude/claude_desktop_config.json on macOS. Developers running from source can install dependencies, build the project, and point the configuration to the generated dist/main.js file.

By default, job data is stored at ~/.job-search-mcp/jobs.json for an installed package or ./data/jobs.json when running from source. JOB_SEARCH_MCP_DATA can override that location.

Tools and capabilities

The server exposes tools for:

• Searching jobs with role, location, and optional source-specific filters.
• Scoring search results and rescoring the current board.
• Showing the ranked board or the saved tracker.
• Applying, skipping, or saving one job.
• Bulk triage by score or source.
• Listing promising and unscored jobs as text.
• Clearing unscored entries or the complete board without removing applied or dismissed records.
• Saving the resume-derived profile used for searches and scoring.

Limitations and notes

Servation/job-search-mcp MCP server is built and tested for Claude Desktop. It should also work with MCP clients that support local stdio servers and MCP App UIs, including Cursor, but other listed examples are untested. Remote-only clients such as Claude on the web or mobile cannot launch the local server.

LinkedIn uses public guest endpoints by default. An optional logged-in mode accepts LINKEDIN_LI_AT and LINKEDIN_JSESSIONID, but the README warns that automated logged-in access may violate LinkedIn’s terms, can lead to account restrictions, and depends on cookies and internal endpoints that may expire or change. If that mode fails, the server falls back to guest access.

Job-source availability can change, and company slugs may become outdated. The repository includes a canary command for checking whether sources still return postings. No account or API key is required for the default setup.