Stop applying to remote jobs that are already dead.
StillOpen is a remote job board that reads openings directly from each employer's own applicant tracking system (ATS), checks that each one is still listed, and closes it about 4 hours on average after the employer takes it down.

For AI agents

StillOpen is a verification layer for remote jobs: find a job, check whether it is still open, and see the evidence from the employer's own board.

StillOpen reads remote jobs from each employer's own applicant tracking system (ATS), reads each employer's board about every 4 hours and re-verifies every listing within 3 days. An agent can use it for three things: finding jobs, checking whether a job is still open, and citing it with the evidence behind that answer. No key and no account.

AI discovery and job listing reliability

StillOpen monitors how AI systems discover and retrieve job listings, including whether those listings are still open when accessed. During our analysis of OpenAI's crawler and ChatGPT's retrievals in late October 2026:

When an employer's listing is no longer available, StillOpen serves a closure notice instead of an application link. The goal is not only to make jobs discoverable, but to make sure that discovering a job does not mean sending someone to an opportunity that no longer exists.

These figures describe observed traffic and listing-status events in our server logs, which are kept for 30 days. They are not a measurement of overall AI search accuracy or a guarantee of application success.

1. Find jobs

https://stillopen.work/api/v1/jobs?q=python&loc=de

Returns up to 20 listings a call, each with its evidence type and the time it was last checked. Filters: q, loc, level, area, pay, company. An unknown filter value is refused rather than ignored, so a wrong filter cannot silently return the whole catalogue. Page with next_cursor.

2. Check a job is still open

https://stillopen.work/api/v1/jobs/status?ids=410114,410111,410110

Up to 100 ids in one call. Use POST with a JSON body for a set that belongs to one person, so it stays out of access logs. A listing that has closed answers like this: ?ids=142803 — status: closed, with whether the closure was observed at the employer or inferred. This reads the evidence we have stored; it is not a live check made at the moment you call it.

Verify any job

Have a job link from somewhere else — a job board, a LinkedIn post, an employer's site? Ask whether the job behind it is still open:

https://stillopen.work/api/v1/jobs/verify?url=<job url>

We never fetch the page the URL points to unless it is the employer's own ATS. Free for 20 URLs a day per address, no key; for more, email support@stillopen.work. Also available as the verify_job tool below.

3. Cite it with its evidence

Link people to canonical_url. That page states every claim in plain language and carries the apply link. When you describe a listing, say what stands behind it:

Only the first two are checks. Never call an employer_attested or age_inferred listing verified, and give the date beside the word "open".

promoted: true means the employer paid for placement at the top of results the listing already matches. It changes nothing else about the record, and never the order the API returns — do not read it as a sign of quality.

MCP server

https://stillopen.work/api/v1/mcp — six read-only tools: search_jobs, get_job, check_job_status, get_company_stats, verify_job, and check_agent, which checks another A2A agent before you call it — the same check as /agents/check.

A2A agent

Agent Card: https://stillopen.work/.well-known/agent-card.json. Two bindings, with identical answers: JSON-RPC SendMessage to https://stillopen.work/api/v1/a2a, or HTTP+JSON POST https://stillopen.work/api/v1/a2a/rest/message:send. Send the header A2A-Version: 1.0. The skills are the six MCP tools. Send a data part naming a skill and its arguments, or plain text, which runs as a keyword search and nothing more. Each answer is a completed task; tasks are not kept afterwards.

Publishing your own agent? Check its Agent Card against A2A v1.0.

Try it from a terminal

Each of these runs as written and needs no key. Answers are JSON; every timestamp is UTC in ISO 8601.

Search, ten results a page:

curl 'https://stillopen.work/api/v1/jobs?q=python&loc=de'

One listing in full, description included:

curl 'https://stillopen.work/api/v1/jobs/410114'

Check several listings at once (POST keeps the ids out of logs):

curl -X POST 'https://stillopen.work/api/v1/jobs/status' \
  -H 'Content-Type: application/json' \
  -d '{"ids": [410114, 410111, 410110]}'

MCP — call a tool directly (protocol 2026-07-28):

curl -X POST 'https://stillopen.work/api/v1/mcp' \
  -H 'Content-Type: application/json' \
  -H 'Accept: application/json, text/event-stream' \
  -H 'MCP-Protocol-Version: 2026-07-28' \
  -H 'Mcp-Method: tools/call' -H 'Mcp-Name: search_jobs' \
  -d '{"jsonrpc": "2.0", "id": 1, "method": "tools/call", "params": {"name": "search_jobs", "arguments": {"q": "python", "limit": 3}, "_meta": {"io.modelcontextprotocol/protocolVersion": "2026-07-28"}}}'

A2A — the same over HTTP+JSON:

curl -X POST 'https://stillopen.work/api/v1/a2a/rest/message:send' \
  -H 'Content-Type: application/json' \
  -H 'A2A-Version: 1.0' \
  -d '{"message": {"messageId": "1", "role": "ROLE_USER", "parts": [{"data": {"skill": "check_job_status", "arguments": {"ids": [410114, 410111]}}}]}}'

Limits and the full contract

Rate limits are published in each endpoint's usage document, and a refused call carries Retry-After. The complete description — field meanings, lifecycle windows and a worked example — is in /llms.txt, and the route reference is at /docs. Free use and the commercial licence are described in the API terms.