Ollama MCP Server

An enhanced MCP (Model Context Protocol) server for interacting with the Ollama API, providing a robust bridge between Claude Desktop and locally-running LLMs via Ollama.

Features

• ✅ Complete Ollama API Coverage: All major Ollama endpoints implemented
• 🔄 Connection Pooling: Efficient HTTP client with connection reuse
• 🚦 Smart Retry Logic: Automatic retries with exponential backoff
• 📝 Comprehensive Logging: Configurable logging for debugging
• ⚡ Response Caching: Intelligent caching for improved performance
• 🛡️ Error Handling: Graceful error handling with helpful messages
• ⚙️ Flexible Configuration: Environment variables and .env file support
• 🌐 Smart Host Detection: Automatically detects localhost vs external network access
• 🔍 Type Safety: Full Pydantic validation for requests/responses
• 📊 Advanced Options: Support for temperature, top_p, seed, and more
• 🌊 Streaming Support: Real-time token streaming for long responses

Quick Start

Smithery Integration

npx -y @smithery/cli@latest install @cuba6112/ollama-mcp --client windsurf --key YOUR_KEY

Prerequisites

• Python 3.9+ installed.
• Ollama is installed and running on your local machine. You can download it from ollama.com.
• uv or pip is installed for package management.

Installation

1. Navigate to the project directory:

   cd /Users/mac_orion/mcp_server/ollama_mcp_server

2. Create a virtual environment:

   Using venv:

   python -m venv .venv
   source .venv/bin/activate

3. Install dependencies:

   Using pip:

   pip install -e . 

   Or using uv:

   uv pip install -e .

Running the Server

Once the dependencies are installed, you can run the server directly:

python -m ollama_mcp_server.main

Or use the mcp dev tool for development:

mcp dev ollama_mcp_server/main.py

Claude Desktop Configuration

To use this server with the Claude Desktop app, you need to add it to your configuration file.

On macOS, edit ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "ollama": {
      "command": "/path/to/your/venv/bin/python",
      "args": ["-m", "ollama_mcp_server.main"],
      "cwd": "/Users/mac_orion/mcp_server/ollama_mcp_server"
    }
  }
}

Replace /path/to/your/venv/bin/python with the actual path to your Python executable.

Available Tools

Core Tools

• list_models: List all available Ollama models with size and modification info
• show_model: Get detailed information about a specific model
• check_model_exists: Check if a model exists locally

Generation Tools

• generate_completion: Generate text completions with advanced options
  • Supports temperature, top_p, top_k, seed, num_predict, stop sequences
  • Streaming support for real-time responses
• generate_chat_completion: Generate chat responses with conversation history
  • Full message history support (system, user, assistant roles)
  • Same advanced options as completion
• generate_embeddings: Create embeddings for text (supports both single strings and lists)

Model Management

• pull_model: Download models from the Ollama library
• copy_model: Duplicate a model with a new name
• delete_model: Remove models from local storage
• list_running_models: Show currently loaded models in memory

Configuration

The server can be configured using environment variables or a .env file:

Connection Settings

# Ollama host - automatically detected by default
OLLAMA_HOST=http://localhost:11434      # Manual override for Ollama API URL
OLLAMA_REQUEST_TIMEOUT=30.0             # Request timeout in seconds
OLLAMA_CONNECTION_TIMEOUT=5.0           # Connection timeout in seconds
OLLAMA_MAX_RETRIES=3                    # Max retry attempts
OLLAMA_RETRY_DELAY=1.0                  # Initial retry delay

Host Auto-Detection

The server automatically detects the appropriate Ollama host:

1. Environment Variable: If OLLAMA_HOST is set, uses that value
2. Localhost Test: Tries to connect to http://localhost:11434
3. Network Detection: If localhost fails, automatically detects local network IP
4. Fallback: Uses localhost as final fallback

This ensures seamless operation in both local development and external access scenarios (like Smithery).

Other Settings

# Logging
OLLAMA_LOG_LEVEL=INFO                   # Log level (DEBUG, INFO, WARNING, ERROR)
OLLAMA_LOG_REQUESTS=false               # Log all API requests/responses

# Performance
OLLAMA_ENABLE_CACHE=true                # Enable response caching
OLLAMA_CACHE_TTL=300                    # Cache TTL in seconds

Copy .env.example to .env and customize as needed.

Troubleshooting

MCP Server Not Connecting

If Claude Desktop shows connection errors:

1. Restart Claude Desktop after making configuration changes
2. Check Ollama is running: ollama ps should show running models
3. Verify Python path in your Claude Desktop config is correct
4. Check logs by setting OLLAMA_LOG_LEVEL=DEBUG in your .env file

Schema Mismatch Errors

If you see parameter-related errors:

1. Restart the MCP server completely
2. Restart Claude Desktop
3. Check that all dependencies are installed: pip install -r requirements.txt

Connection to Ollama Fails

If the server can't connect to Ollama:

1. Ensure Ollama is running: ollama serve
2. Check the Ollama URL in your configuration
3. Try accessing Ollama directly: curl http://localhost:11434/api/tags

Performance Issues

For better performance:

1. Enable caching: OLLAMA_ENABLE_CACHE=true
2. Adjust cache TTL: OLLAMA_CACHE_TTL=600
3. Increase timeout for large models: OLLAMA_REQUEST_TIMEOUT=60

Testing

Test the server directly using the MCP dev tool:

mcp dev ollama_mcp_server/main.py

Or run the server and test individual tools:

# Start the server
python -m ollama_mcp_server.main

# In another terminal, test with Claude Desktop or other MCP clients
# You can also check the example usage:
python examples/usage_example.py

This will verify that all core functions work correctly with your Ollama installation.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request. For major changes, please open an issue first to discuss what you would like to change.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• Anthropic for the Model Context Protocol specification
• Ollama for the excellent local LLM platform
• The MCP community for tools and documentation

Support

If you encounter any issues or have questions:

• Check the troubleshooting section above
• Look through existing GitHub issues
• Create a new issue with detailed information about your problem