Ask Ostra Pro

The ostra_pro_chat tool on the Ostra MCP server — arguments, result, errors and examples.

MCPhttps://ostra.testrelic.ai/mcp · ostra_pro_chat

Asks Ostra Pro, TestRelic’s testing model, a question: write or fix a test, analyse a failure or log, read a screenshot.

Send either prompt (with optional system and images) or a full messages array, not both. Long answers can stream progress: see MCP overview.

Title
Ostra Pro: ask the testing model
Access
Full-access key. Read-only keys can list the tool but not call it.
Cost
Per token, the same as chat completions.
Hints
readOnlyHint: true · destructiveHint: false · idempotentHint: false · openWorldHint: false

Once the Ostra MCP server is connected, your agent calls the tool with these arguments.

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "ostra_pro_chat",
    "arguments": {
      "prompt": "This Playwright test times out on CI but passes locally. Why, and how do I fix it?",
      "reasoning_effort": "low",
      "max_tokens": 2048,
      "session": "checkout-fix-1"
    }
  }
}
Headers
mcp-protocol-version
2025-06-18
{
  "jsonrpc": "2.0",
  "id": 1,
  "result": {
    "content": [
      {
        "type": "text",
        "text": "CI machines are slower and run tests in parallel, so the coupon button renders after your 30 s timeout. Wait for the state you need instead of a fixed time: `await expect(page.getByRole('button', { name: 'Apply coupon' })).toBeEnabled()` before clicking, and stop sharing the cart between tests so parallel workers don’t race."
      }
    ],
    "structuredContent": {
      "model": "ostra-pro",
      "version": "1.0",
      "content": "CI machines are slower and run tests in parallel, so the coupon button renders after your 30 s timeout. Wait for the state you need instead of a fixed time: `await expect(page.getByRole('button', { name: 'Apply coupon' })).toBeEnabled()` before clicking, and stop sharing the cart between tests so parallel workers don’t race.",
      "finish_reason": "stop",
      "usage": {
        "input_tokens": 61,
        "cached_input_tokens": 0,
        "output_tokens": 170
      },
      "cost_usd": "0.00079894",
      "balance_usd": "24.99920106",
      "request_id": "req_01JDXKB3C4D5E6F7G8H9J0K1L2"
    },
    "isError": false
  }
}

Arguments

  • promptstring

    The question or task. Use this or messages.

  • systemstring

    System prompt to go with prompt.

  • imagesarray of strings

    Image URLs to go with prompt: https://… or data:image/…;base64,….

    0–8 items
  • messagesarray of objects

    A full OpenAI-style conversation instead of prompt/system/images.

  • max_tokensinteger

    Most output tokens, reasoning included.

    1 to 32,768default 8192
  • reasoning_effortstring

    How long to think; none turns thinking off.

    One ofnonelowmediumxhighdefault "xhigh"
  • temperaturenumber

    Sampling temperature.

    0 to 2
  • sessionstring

    Optional conversation id, one per conversation or agent run (like x-ostra-session). Calls with the same value go to the same model replica when possible (best effort), so follow-up calls more often bill a repeated prompt prefix as cached input.

    pattern ^[\w.:-]{1,128}$

Result

content[0].text holds the answer as text; structuredContent holds it as JSON:

  • modelstring

    ostra-pro.

  • versionstring

    1.0.

  • contentstring

    The answer (also the tool result’s text content).

  • reasoning_contentstring

    The reasoning, when thinking was on.

  • finish_reasonstring

    stop, length or tool_calls.

  • usageobject

    input_tokens, cached_input_tokens, output_tokens.

  • cost_usdstring

    What the call cost.

  • balance_usdstring

    Balance after the call.

  • request_idstring

    Request id.

Errors

Billing, validation, rate-limit and permission problems come back as a normal result with isError: true and structuredContent.error (type, message), the same types as the REST errors. Authentication failures are HTTP 401/403 before any JSON-RPC.

Was this page helpful?