jobs.ch MCP server: let your AI agent search Swiss job ads

#What your agent gets

The tool takes a required term, for example python, pflegefachfrau or buchhaltung. The optional inputs are location and page. location takes a city, canton or postcode, such as Zürich, Bern or 8001. page defaults to 1.

#Inputs and output

The result has term, location, total, page and jobs. total is the number of matches jobs.ch reports for the search, and it can be null when the page does not show one. Each job has title, company, place, employment_type, posted_at and url. The title can include the workload, such as 80–100 %.

#How a request flows

1. You ask your agent: "Find Python jobs in Zürich on jobs.ch, with title, company and link."
2. The agent calls find_recipe, then run with jobs.ch/search-jobs, term set to python and location set to Zürich.
3. wield-mcp passes the run through the local bridge to the Wield extension.
4. The extension sends the search request from your browser, so the recipe runs in your own browser and not on a server.
5. The job ads go back to your agent. Nothing passes through wield.run.

#What we saw in our check

On 7 November 2026, a search for python in Zürich passed all 5 checks in one request, in 548 ms, and returned 121 matches. A search for verkauf on page 2 returned 3,245 matches, which the recipe read correctly from the page. A made-up term does not return zero results. jobs.ch's search is fuzzy, so a nonsense term still found about 112 loosely related jobs. Read the titles before you rely on a result.

The recipe reads the job list that the results page publishes as structured data. That list holds the page's 20 ads, each with its title, company, place, employment type, publication date and link.

#What leaves your computer

The recipe makes one request, to www.jobs.ch, for one results page, as the site serves it to a visitor. The request leaves from your browser. Nothing passes through wield.run. Wield logs each request on your computer only, with method, host, path, status and timing, and never logs bodies or query values. Telemetry is off by default. Wield never reads, copies or sends your cookies or passwords.

#jobs.ch's terms, in plain words

The site's robots.txt allows the search page. It disallows /api/, login and feature-flag URLs. It has no rules for Claude agents or AI content signals. The recipe makes one request per call. It does not submit applications, set up job alerts or save bookmarks, and it makes no purchases. Our review of jobs.ch's terms is not finished, and this is not legal advice. Use the recipe for one search that you start yourself, in your own browser, and not for bulk collection of ads. Read the site's terms yourself before you use this recipe in any other way.

#Limits and what is not working yet

• Read only. The recipe searches. It does not apply, set up job alerts or save bookmarks.
• Fuzzy matches. A term with no real match still returns loosely related jobs.
• One page per call. Each call returns about 20 jobs. Pass page to read the next page.
• Nullable fields. location and total can be null.
• Run limits apply. A run has at most 12 steps, 40 requests, 5 MB per response and 60 seconds.

#FAQ

#Can the agent apply for a job?

No. The recipe only searches. Open the ad's url to apply on jobs.ch yourself.

#Does it need a jobs.ch login?

No. The search works without an account.

#Why do unrelated jobs show up for a made-up word?

jobs.ch's search is fuzzy. In our check, a nonsense term still returned about 112 loosely related jobs, so read the titles before you rely on a result.

#Which AI apps work with it?

Any MCP client. npx -y wield-mcp init sets up Claude Desktop, Claude Code, Cursor and Codex. Other MCP clients work with the standard stdio config.

#Where do the results go?

To your agent, on your own computer. The results never pass through wield.run.