API reference¶

The bridge listens on http://localhost:8000 by default and mirrors the Ollama API. Interactive Swagger UI is at /docs while it's running.

Endpoints¶

`POST /api/chat`¶

Ollama's chat endpoint, plus MCP tools. This is the only endpoint where tools are integrated; the tool-calling loop runs server-side and the client sees a single request/response exchange.

Accepts the same payload as Ollama, including stream, think and options.

curl -N -X POST http://localhost:8000/api/chat \
  -H "accept: application/json" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen3:0.6b",
    "messages": [
      {"role": "system", "content": "You are a weather assistant."},
      {"role": "user", "content": "What is the weather like in Paris today?"}
    ],
    "think": true,
    "stream": true,
    "options": {"temperature": 0.7, "top_p": 0.9}
  }'

`GET /health`¶

Bridge health and status. Not proxied to Ollama.

curl http://localhost:8000/health

`GET /version`¶

Installed version plus the result of the PyPI update check. Not proxied.

curl http://localhost:8000/version

Everything else¶

Every other path and method is proxied to Ollama verbatim — /api/generate, /api/tags, /api/embed, /api/pull, and anything Ollama adds later. The catch-all route matches all methods, so no endpoint list needs maintaining.

Only /api/chat gets tools

/health and /version belong to the bridge. Everything else is a transparent passthrough. If you need MCP tools, you need /api/chat.

Using it as a drop-in¶

The bridge is a drop-in proxy for the Ollama API. Existing clients and libraries keep working against both local and cloud Ollama models — change the host and nothing else.

from ollama import Client

client = Client(host="http://localhost:8000")
client.chat(model="qwen3", messages=[...])

Tip

/docs gives you the whole surface interactively, which is usually faster than reading this page.