Skip to main content
Glama
Job Radar

Job Radar

README.md
# Job Radar

An MCP server that gives AI agents job-search tools. It sweeps company job boards on Greenhouse, Lever and Ashby through their public posting APIs: no scraping, no login.

17 boards, about 2,600 postings per sweep, in a few seconds. One file, standard library plus the `mcp` package.

## Tools

| Tool | What it does |
|---|---|
| `find_jobs(want, skip, location)` | Sweeps every board, filters by title words, hides jobs you already applied to |
| `mark_applied(url)` | Records an application so the job stops showing up |
| `applied_list()` | Lists everything marked as applied |

The defaults fit about 2 years of .NET/full-stack experience: `want` matches .NET, backend and full-stack titles; `skip` drops senior, staff, lead and manager roles. Pass your own words to change that, or edit `WANT` and `SKIP` in `server.py`.

## Install

Needs Python 3.10+.

```bash
pip install -r requirements.txt
```

Register it with Claude Code (use the full path to `server.py`):

```bash
claude mcp add job-radar -- python /path/to/job-radar/server.py
```

Then ask in any session: "what's new on job radar in India?"

Any other MCP client works the same way: run `python server.py` over stdio.

## Without an agent

```bash
python server.py --html India
```

Writes `jobs.html` with the current matches for that location and opens it.

## Self-check

```bash
python server.py --check
```

Asserts the title filter, then calls every board live and prints which are alive and which are dead. Run it after editing `watchlist.json`, because board tokens go stale.

## Adding companies

Add the company's token to `watchlist.json` under its ATS, then run `--check`. The token is the slug in the careers URL: `job-boards.greenhouse.io/gitlab` gives `gitlab`.

## Limits

- Only companies on Greenhouse, Lever or Ashby. Many Indian companies use Darwinbox, Keka or Workday, which have no public posting API.
- Filtering is on the job title only, not the description.
- Applications are tracked in a local `applied.json`.

## License

MIT