---
title: "Klaro for AI agents: MCP, JSON search, feed and Markdown · Klaro"
description: "How AI assistants and scripts read Klaro's doctor jobs in Germany: MCP server, public JSON search, feed, Markdown copies and llms.txt. No account, no key, read only."
canonical: "https://klaro.tools/for-agents"
language: "en"
date_modified: "2026-11-21"
---

# Klaro for AI agents

Klaro lists the open jobs for doctors in Germany, read from employers' career sites and the Federal Employment Agency. AI assistants and scripts can read this data in six ways. All are free, need no account and no key, and none can change anything.

Addresses and limits checked on 21 November 2026.

## MCP server

For assistants that connect to MCP servers (Model Context Protocol): Claude, ChatGPT and others. Streamable HTTP, no sign-in.

- **Address** `https://klaro.tools/mcp`: Add it to any MCP client as a remote server.
- **[In Claude](https://claude.ai/directory/connectors/klaro-doctor-jobs)**: Listed in the connector directory as "Klaro Doctor Jobs Germany" since 18 November 2026.
- **MCP Registry** `tools.klaro/jobs`: The entry in the official MCP Registry.
- **Tools**: search\_doctor\_jobs, get\_job\_details, compare\_jobs, list\_medical\_specialties, suggest\_job\_locations, search, fetch. All read only: search with 13 filters, one job as Markdown, up to 5 jobs compared, the filter values with counts, town suggestions.
- **Limits**: 20 jobs an answer, 60 tool calls a minute per address, requests up to 64 KB.
- **[Guide](https://klaro.tools/ratgeber/find-doctor-jobs-with-claude)**: Set up Klaro in Claude.

## JSON search

For anything that can open an address. The same filters and meaning as search\_doctor\_jobs on the MCP server.

- **Search** `https://klaro.tools/api/jobs/search?specialty=Kardiologie&state=Hessen`: 30 jobs a page with title, employer, town, state, specialty, career level, date, Klaro page and flags (instant apply, Berufserlaubnis, training authorisation, part time, fixed term). "next" is the next page, up to page 34.
- **Filters**: specialty, state, career\_level, hospital\_type, employer, city, text, contract, posted\_within, instant\_apply\_only, training\_only, berufserlaubnis\_only, sort, page. Values are German ("Innere Medizin", "Bayern", "Assistenzarzt"); the /stellen words work too: fachrichtung, bundesland, stufe, ort, q.
- **Filter values** `https://klaro.tools/api/jobs/facets`: Every value with its number of open jobs and its search address; with filters, the counts within that choice.
- **No filter**: The jobs of the last 30 days, newest first, as on the MCP server.

## Feed

The newest jobs as JSON Feed 1.1 and RSS 2.0, renewed hourly.

- **JSON Feed** `https://klaro.tools/feed/arztstellen.json?fachrichtung=Kardiologie&bundesland=Berlin`: 20 jobs a page, ?page= up to 11, the next page in next\_url.
- **RSS** `https://klaro.tools/feed/arztstellen.xml`: The same jobs; the next page as atom:link rel="next".
- **[For partners](https://klaro.tools/partner)**: Feed addresses by region, widget and logo (German).

## Markdown copies

The public pages as Markdown, for readers that run no JavaScript. The pages themselves show their content without JavaScript too.

- **Add .md** `https://klaro.tools/arztstellen/fach/kardiologie/hessen.md`: Overviews, jobs, hospitals, training atlas, Facharzt pages, guides, blog, report and this page.
- **Accept header** `Accept: text/markdown`: The homepage, /stellen, overviews, jobs, hospitals, Facharzt pages, the report and this page then answer Markdown at the same address. The HTML pages name their copy in their Link header (rel="alternate").
- **Homepage** `https://klaro.tools/index.md`: The homepage too, and /stellen, the calculators, /weiterbildung, /partner and /preise.

## llms.txt

The site's overview for language models.

- **llms.txt** `https://klaro.tools/llms.txt`: What Klaro is, every page type with example addresses, and the ways in on this page. The homepage, guides, blog and calculators point to it in their Link header with rel="describedby".
- **Short answers** `https://klaro.tools/ratgeber/llms.txt`: Every guide, Facharzt page, calculator and 40 blog posts, each with a one or two sentence answer, date and Markdown address.

## WebMCP in the browser

In a browser with WebMCP, every klaro.tools page offers the agent in that tab the same tools.

- **Tools**: search\_doctor\_jobs, get\_job\_details, compare\_jobs, list\_medical\_specialties, suggest\_job\_locations. They answer through the MCP server; nothing changes the page or reads the visitor's data.

## What Klaro never does

- Repeat the text of a job advert. Klaro gives facts and its own summary; the employer's original advert, linked from every job page, is binding.
- Send an application unless the doctor checks it and sends it themselves. None of the ways in on this page can apply.
- Give out application addresses, contact persons or users' data.
- Ask for a sign-in, a key or payment for these ways in.
- Keep what a call asked. The MCP server counts only tool, outcome and duration per day, with no address and no search words.

## How to cite Klaro

Please link to the Klaro page you quote.

- Overviews and Facharzt pages name the day of their count. Quote the number and the date together.
- Finished report months have fixed addresses and no longer change, for example /report/doctor-jobs-germany/2026-10.
- Report numbers and charts are CC BY 4.0, credit "Source: Klaro (klaro.tools)".
- For a single job: the Klaro page and the day you read it.

`Klaro Doctor Jobs in Germany report, October 2026, klaro.tools/report/doctor-jobs-germany/2026-10`

## FAQ

### Do I need a key or an account?

No. The MCP server, the JSON search, the feed and the Markdown copies are free and need no sign-in.

### Can an agent apply for a job through Klaro?

No. Every way in on this page only reads. An application through Klaro is sent by the doctor, signed in on klaro.tools.

### How fresh are the jobs?

Klaro reads employers' career sites several times a day. The JSON search is cached for up to five minutes, the feed for up to an hour.

### May I use the data to train AI models?

Yes. robots.txt allows search, use as a source for answers and training AI models (Content-Signal: search=yes, ai-input=yes, ai-train=yes). Please name Klaro (klaro.tools) as the source.

Deutsch: https://klaro.tools/fuer-agenten

Canonical page: https://klaro.tools/for-agents
