Ollama Mcpo Adapter

🚀 🦙 Ollama-MCP Adapter

Expose MCPO (MCP to OpenAPI proxy server) tools as Ollama-compatible functions using a simple Python adapter and an optional runtime service.

✨ Features

• 🔌 Connect to an MCPO instance.
• ⚙️ Start your own MCPO server via the MCPOService program.
• 🔁 List MCP tools exposed via OpenAPI as Ollama-compatible tool functions.

🚀 Quick Start

1. Installation

pip install ollama-mcpo-adapter

Using an Existing MCPO Instance

Assume you have an MCPO instance running like this:

uvx mcpo --port 5090 --config /path/to/mcp_config.json

You can use the adapter to get all available functions:

from ollama_mcpo_adapter import OllamaMCPOAdapter

adapter = OllamaMCPOAdapter(host="localhost", port=5090, config_path="/path/to/mcp_config.json")
# Get tool descriptions from MCPO FastAPI's /docs
tools = adapter.list_tools_ollama()

You can omit the configuration file path. However, it's more reliable to discover the MCP server name by providing the configuration file. Otherwise, the server name in the OpenAPI documentation automatically generated by MCPO may change.

Using a Local MCPO Service

You can use this extension to start an MCPO service:

from ollama_mcpo_adapter import MCPOService

# Provide an MCP configuration file or dictionary
mcp_config = {
    "mcpServers": {
        "time": {"command": "uvx", "args": ["mcp-server-time", "--local-timezone=Europe/Berlin"]}
    }
}
mcpo = MCPOService("127.0.0.1", 4090, config=mcp_config,
                   # -OR- Load from an existing MCP configuration file
                   config_path="path/to/mcp_config.json")
# The MCPOSService class handles the startup and shutdown of the MCPO server and runs it in a subprocess
mcpo.start(wait=True)
...
mcpo.stop()

Then use the adapter to get all available tools:

from ollama_mcpo_adapter import OllamaMCPOAdapter
adapter = OllamaMCPOAdapter("127.0.0.1", 4090)
tools = adapter.list_tools_ollama()

Send these to Ollama:

from ollama import Client
# Send a prompt to Ollama with the discovered tools
client = Client(host="http://127.0.0.1:11434")
response = client.chat(
    model="qwen2.5-coder:14b-instruct-q4_K_M",
    messages=[{"role": "user", "content": "Write a file..."}],
    tools=tools,
)

Finally, call the tools:

# Handle any tool calls
if response.message.tool_calls:
    adapter.call_tools_from_response(response.message.tool_calls)

Environment Variables

• The path to npx on Windows. You can override it via the configuration parser. The default path is WIN_NODEJS_NPX_PATH=C:\Program Files\nodejs\npx.cmd

🧪 Running Tests

pytest

📂 Project Structure

ollama_mcpo_adapter/
├── adapter.py        # Tool discovery + Ollama integration
├── service.py        # Optional: Program to start MCPO
├── service_runner.py # MCPO subprocess control
├── config_parser.py  # MCP configuration parsing helper
├── dispatcher.py     # Dispatch tool calls

📜 License

MIT. See LICENSE.