Openings MCP server

Openings

A free, candidate-safe job search for AI agents.

Openings indexes public company job boards into a private index on your machine and exposes it to any MCP client. Your agent can find roles that fit a resume, explain the fit with evidence, and propose truthful resume improvements. Installed this way there are no accounts and no API keys; there are never model calls and never a way to submit an application.

Two ways to run it, with different data rules:

Both are covered in Sharing crawls and usage and on the privacy page.

• Eleven providers plus company sites. Greenhouse, Lever, Ashby, Workday, Recruitee, SmartRecruiters, Workable, Breezy, Freshteam, Keka, and Zoho Recruit, crawled from their public structured endpoints, plus employer career sites read only through the schema.org JobPosting markup they publish for search engines. No free-form HTML scraping.
• Verified sources only. Every company in the catalog passed an identity check against its own board.
• Resumes stay in memory. Resume content is parsed for a single request and never written to disk.
• Honest coverage. Before you share a resume, Openings tells you how many employers and jobs it actually has for your countries.

For job seekers

Install Bun 1.3 or newer, then the package with whichever package manager you use:

bun add --global openings
# or
npm install --global openings
# or
yarn global add openings

Bun must be on your PATH either way; it is the runtime the server runs on.

Add the server to your MCP client:

{
  "mcpServers": {
    "openings": { "command": "openings-mcp", "args": [] }
  }
}

Then ask your agent something like:

Show me what Openings covers in India. If that looks useful, use my resume to find backend roles, including good jobs whose titles I would not have searched for. Rank by evidence and explain every gap.

On first use the agent downloads the shared index of every verified source in one call, reports real coverage, and only then asks for a resume; only missing or stale sources are crawled, in batches of 25. Results come back in three buckets: direct title matches, hidden roles found through grounded title families, and stretch roles, each with separate evidence and keyword scores. The index lives under ~/.openings.

The job-seeker quickstart has sample prompts, an example conversation, privacy details, and common errors.

The tools

Shortened. The full README is on GitHub.