Prompts and tests
Prompts
Section titled “Prompts”The server also offers three prompts: ready-made instructions that clients show as commands. In Claude Code they are /mcp__kurlo__sync_collection and so on.
| Prompt | Arguments | What the assistant does |
|---|---|---|
sync_collection | collection, source | Finds the HTTP routes in your project’s code (Express, FastAPI, Flask, Go routers, Spring, Rails, ASP.NET and others), matches them with the saved requests, and adds a request with tests for each route that has none. It deletes nothing: saved requests that match no route are listed for you to decide. |
cover_endpoint | endpoint (required), collection | Does the same for one endpoint, such as POST /orders, and also saves its main error responses as examples. |
debug_last_failure | entry | Calls diagnose_last_failure, explains the cause and proposes a fix, which it applies only after you agree. |
Requests are saved with {{baseUrl}} in the URL, so the address lives in an environment, not in each request. When no environment sets it, the prompts tell the assistant to look for it in the project (docker-compose.yml, launch settings, the README) and to ask you before calling set_variable, and never to point an environment at production. With --read-only only debug_last_failure is offered.
Tests from a response
Section titled “Tests from a response”generate_tests writes a saved request’s tests for it. With from: run, the default, it sends the request and writes three pm.test checks from the response:
- the status, such as
pm.response.to.have.status(200); - the
Content-Type; - a JSON schema of the body, checked with
pm.response.to.have.jsonSchema: the type of every field and which fields are required. Values are left out, so the tests do not break when the next order has another id, and nothing from the response is written into the request’s file.
The response is also saved as an example, named after its status, so the mock server serves it. Run generate_tests again later and an example of the same name is refreshed rather than added twice.
Sending the request has its effects: a POST creates what it creates. A response with a 4xx or 5xx status is refused, since the tests would assert a failure, unless the call passes allowErrorStatus: true to test that error on purpose.
When the API is not running anywhere yet, from: example takes what the code returns instead: a status and a body, or a schema built from the response type. The tests and the example are written the same way, and the tests pass once the server is up. A schema that does not accept its own body is refused.
Tests the request already has are kept and the new ones are added after them. A test with the same name, such as status is 200, is not added twice. When the request’s tests already expect another status, an error response such as a 422 is saved as an example only, so the script never asserts two statuses. replace: true replaces the whole test script.