mcp-ollama-agent

This project demonstrates integration between Model Context Protocol (MCP) servers and Ollama, allowing AI models to interact with various tools through a unified interface.

✨ Features

• Supports multiple MCP servers (both uvx and npx tested)
• Built-in support for file system operations and web research
• Easy configuration through mcp-config.json similar to claude_desktop_config.json
• Interactive chat interface with Ollama integration that should support any tools
• Standalone demo mode for testing web and filesystem tools without an LLM

🚀 Getting Started

1. Prerequisites:

2. Node.js (version 18 or higher)

3. Ollama installed and running

4. Install the MCP tools globally that you want to use:

   ```bash # For filesystem operations npm install -g @modelcontextprotocol/server-filesystem

   # For web research npm install -g @mzxrai/mcp-webresearch ```

5. Clone and install:

```bash git clone https://github.com/ausboss/mcp-ollama-agent.git cd mcp-ollama-agent npm install

```

1. Configure your tools and tool supported Ollama model in mcp-config.json:

json { "mcpServers": { "filesystem": { "command": "npx", "args": ["@modelcontextprotocol/server-filesystem", "./"] }, "webresearch": { "command": "npx", "args": ["-y", "@mzxrai/mcp-webresearch"] } }, "ollama": { "host": "http://localhost:11434", "model": "qwen2.5:latest" } }

1. Run the demo to test filesystem and webresearch tools without an LLM:

bash npx tsx ./src/demo.ts

1. Or start the chat interface with Ollama: bash npm start

⚙️ Configuration

• MCP Servers: Add any MCP-compatible server to the mcpServers section
• Ollama: Configure host and model (must support function calling)
• Supports both Python (uvx) and Node.js (npx) MCP servers

💡 Example Usage

This example used this model qwen2.5:latest

Chat started. Type "exit" to end the conversation. You: can you use your list directory tool to see whats in test-directory then use your read file tool to read it to me? Model is using tools to help answer... Using tool: list_directory With arguments: { path: 'test-directory' } Tool result: [ { type: 'text', text: '[FILE] test.txt' } ] Assistant: Model is using tools to help answer... Using tool: read_file With arguments: { path: 'test-directory/test.txt' } Tool result: [ { type: 'text', text: 'rosebud' } ] Assistant: The content of the file `test.txt` in the `test-directory` is: rosebud You: thanks Assistant: You're welcome! If you have any other requests or need further assistance, feel free to ask.

System Prompts

Some local models may need help with tool selection. Customize the system prompt in ChatManager.ts to improve tool usage.

🤝 Contributing

Contributions welcome! Feel free to submit issues or pull requests.