chat

Instructions

Implementation Reference

• src/foundry_reverse/server.py:189-219 (registration)

  MCP tool registration and handler function for the 'chat' tool. The @mcp.tool decorator registers it with name='chat' and the async function implements the handler logic, calling oc.chat() to delegate to the Ollama client.

  @mcp.tool(
      name="chat",
      description=(
          "Send a multi-turn conversation to an Ollama model.  Messages should "
          "follow the format [{'role': 'user'|'assistant'|'system', 'content': '...'}]."
      ),
  )
  async def chat(
      model: str,
      messages: list[dict[str, str]],
      temperature: float | None = None,
      max_tokens: int | None = None,
  ) -> dict[str, Any]:
      """
      Args:
          model: Ollama model name.
          messages: Conversation history as a list of role/content dicts.
          temperature: Sampling temperature (0.0–2.0).
          max_tokens: Maximum tokens to generate.
      """
      options: dict[str, Any] = {}
      if temperature is not None:
          options["temperature"] = temperature
      if max_tokens is not None:
          options["num_predict"] = max_tokens
      response = await oc.chat(
          model=model,
          messages=messages,
          options=options or None,
      )
      return {"model": model, "response": response}

• src/foundry_reverse/ollama_client.py:78-93 (helper)

  Helper function that makes the HTTP POST to Ollama's /api/chat endpoint. Constructs the payload with model, messages, and options, then returns the assistant's content from the response.

  async def chat(
      model: str,
      messages: list[dict[str, str]],
      options: dict[str, Any] | None = None,
  ) -> str:
      payload: dict[str, Any] = {
          "model": model,
          "messages": messages,
          "stream": False,
      }
      if options:
          payload["options"] = options
      async with _client() as c:
          r = await c.post("/api/chat", json=payload)
          r.raise_for_status()
          return r.json().get("message", {}).get("content", "")

• tests/test_ollama_client.py:53-68 (helper)

  Unit test for the ollama_client.chat() function, mocking the HTTP client to verify correct request/response behavior.

  @pytest.mark.asyncio
  async def test_chat():
      mock_client = AsyncMock()
      mock_client.__aenter__ = AsyncMock(return_value=mock_client)
      mock_client.__aexit__ = AsyncMock(return_value=False)
      mock_client.post = AsyncMock(
          return_value=_mock_response({"message": {"content": "Hi there!", "role": "assistant"}})
      )
  
      with patch("foundry_reverse.ollama_client.httpx.AsyncClient", return_value=mock_client):
          result = await oc.chat(
              model="llama3",
              messages=[{"role": "user", "content": "Hello"}],
          )
  
      assert result == "Hi there!"

• src/foundry_reverse/server.py:150-151 (schema)

  Comment delineating the INFERENCE section which includes both 'generate' and 'chat' tools.

  # INFERENCE  (generate / chat)
  # ────────────────────────────────────────────────────────────────────────────