An application's message format can tie it to a particular model provider. Ollama's Anthropic-compatible endpoint lets the Python Anthropic client obtain a local model's reply while retaining the Messages API format.

Messages responses contain a list of content blocks rather than a single answer string. Text blocks hold the answer, while other block types can carry reasoning or tool requests; a completed text reply is enough to establish the basic client connection.

Python 3.10 or later with virtual-environment support and a running local Ollama server are required. The server needs an installed chat model such as qwen2.5:0.5b, and the Python process must run on the same machine for a loopback address to reach it. Ollama implements a subset of the Anthropic API, so hosted Claude features are not automatically available.

Steps to use an Anthropic-compatible client with Ollama:

  1. Create a virtual environment for the Anthropic library.
    $ python3 -m venv .venv

    The shell commands use POSIX virtual-environment paths. On Windows, the environment's Python executable is in its Scripts directory.

  2. Install the official Anthropic Python SDK.
    $ .venv/bin/python -m pip install anthropic
  3. Create ollama-anthropic.py with the local Messages client.
    ollama-anthropic.py
    from anthropic import Anthropic
     
    client = Anthropic(
        base_url="http://127.0.0.1:11434",
        api_key="ollama",
    )

    The SDK adds /v1/messages to this server address. The literal ollama key is a placeholder accepted by local Ollama, not a hosted Anthropic credential. It does not protect the server from other callers.
    Related: How to start the Ollama server

  4. Append a question for the installed Ollama model to ollama-anthropic.py.
    message = client.messages.create(
        model="qwen2.5:0.5b",
        max_tokens=80,
        messages=[
            {"role": "user", "content": "What is 17 plus 25? Answer in one short sentence."}
        ],
    )

    The model value is an Ollama tag, not a Claude model name. The token budget allows a short reply. If the model is missing, it needs to be pulled on this server before the request can succeed.
    Related: How to pull an Ollama model

  5. Append text-block extraction and completion checks to ollama-anthropic.py.
    text = "".join(block.text for block in message.content if block.type == "text")
    assert message.stop_reason == "end_turn", message
    assert text.strip(), message
    print(f"Stop reason: {message.stop_reason}")
    print(text)

    The text comes from the returned message object. A missing answer or a stop reason such as max_tokens raises an error instead of passing an incomplete reply.

  6. Verify the completed Messages reply by running ollama-anthropic.py.
    $ .venv/bin/python ollama-anthropic.py
    Stop reason: end_turn
    17 plus 25 equals 42.

    The answer's wording can vary. A connection error means the client cannot reach the server address; a model-not-found error means the requested tag is absent there. The printed stop reason and text show that the SDK received and parsed a completed reply.

Share this guide