What is an MCP job search server, and how does Openings work?
Short answer: the Model Context Protocol lets an AI assistant call outside tools. Openings is an MCP server for job search: seven read-only tools over 469,607 live postings read from employers' own boards, at https://openings.avagama.co/mcp. It ranks roles against a resume with quoted evidence, and it cannot apply on your behalf.
The problem it solves
An assistant asked for jobs has two bad options: refuse, or recall listings it cannot verify. Neither helps someone job hunting. What changes the answer is a source the assistant can query during the conversation, with dates it can quote and links it can hand over.
The seven tools
| Tool | What it does |
|---|---|
| prepare_job_search | Builds or refreshes the index for your countries and reports what it holds |
| get_job_coverage | Says how many employers and roles exist for a country, before you share anything |
| search_jobs | Keyword search by title and employer, with country, city and remote filters |
| recommend_jobs | Ranks roles against a resume, in direct, hidden and stretch groups |
| analyze_job_fit | Explains one role against verbatim resume evidence, including screening risks |
| optimize_resume | Proposes grounded changes and a diff, inventing nothing |
| get_job | Fetches one posting in full, with the employer's application link |
What evidence-based matching means
A recommendation is only as good as its reasons. Each match carries the requirements it satisfied, the ones it satisfies indirectly, the gaps, and the exact resume spans behind them. If a posting states the experience it wants, the range is quoted and compared with what the resume actually shows, rather than guessed from the word "senior" in the title.
Freshness, stated rather than implied
Results start with roles posted in the last seven days and widen to fourteen, then thirty, then everything, only when a week is thin, and the answer says when it widened. Each role carries its age and whether the date came from the employer or from our first sighting. A posting with no published date is labelled as such, never quietly treated as new.
Two ways to run it
| On your machine | Hosted connector | |
|---|---|---|
| Install | bun add --global openings | Paste one address |
| Account | None | Email and a six-digit code |
| Index | Built and stored locally | Shared, refreshed centrally |
| Resume | Parsed in memory, never written to disk | Sent, used for one request, discarded |
What it deliberately will not do
- No applying. Every tool is read-only and says so, which is why an app can be told to always allow them.
- No invented experience. Resume suggestions are grounded in facts already in the document.
- No completeness claim. Coverage is reported before you share a resume, so you can judge the source first.
- No crawling where we are told not to. Employers' robots rules are respected, and boards whose terms forbid automated collection are left out.
index.md, the site index for language models is at avagama.co/llms.txt, and live counts are at summary.json.Common questions
What is MCP?
The Model Context Protocol is an open standard that lets an AI assistant call outside tools and read outside data. An MCP server publishes a set of tools; an app such as Claude, ChatGPT, Cursor or VS Code connects to it and the assistant can then use them during a conversation.
What is an MCP job search server?
It is an MCP server whose tools search job postings. Openings publishes seven: prepare a search for your countries, report coverage, search by keyword, recommend roles against a resume, analyse fit for one role, suggest resume improvements, and fetch one posting in full.
How is that different from a job board?
A job board is a website you read. An MCP server is a source your assistant can query while reasoning, so it can compare a resume against hundreds of postings and explain each match with quoted evidence rather than leaving you to read them all.
Where do the postings come from?
From employers' own boards: Greenhouse, Lever, Ashby, Workday, SmartRecruiters, Workable, Breezy, Freshteam, Keka, Zoho Recruit, Oracle Cloud Recruiting, and company career sites that publish structured job data. Every link opens the employer's own posting.
Is my resume stored anywhere?
No. Run locally, the resume is parsed in memory for one request and never written to disk. Through the hosted connector it is sent to the server, used for that request and discarded. Nothing derived from it is kept except anonymous skill and title words, which you can turn off.
Can it apply to jobs for me?
No, and it is built so it cannot. All seven tools are read-only and declare themselves as such, so an app can auto-approve them safely.