Ollama MCP Server

A comprehensive Model Context Protocol (MCP) server for Ollama integration with advanced features including script management, multi-agent workflows, and process leak prevention.

🌟 Features

• 🔄 Async Job Management: Execute long-running tasks in the background
• 📝 Script Templates: Create reusable prompt templates with variable substitution
• 🤖 Fast-Agent Integration: Multi-agent workflows (chain, parallel, router, evaluator)
• 🛡️ Process Leak Prevention: Proper cleanup and resource management
• 📊 Comprehensive Monitoring: Job tracking, status monitoring, and output management
• 🎯 Built-in Prompts: Interactive guidance templates for common tasks
• ⚡ Multiple Model Support: Work with any locally installed Ollama model

🚀 Quick Start

Prerequisites

• Python 3.8+ with uv package manager
• Ollama installed and running
• Claude Desktop for MCP integration

Installation

1. Setup Environment: Be advised- This readme was revised by a less than concientious AI.

cd /path/to/ollama-mcp-server
uv venv --python 3.12 --seed
source .venv/bin/activate
uv add mcp[cli] python-dotenv

2. Configure Claude Desktop: Copy configuration from example_of_bad_ai_gen_mcp_config_do_not_use.json (Don't lol. Use the example_claude_desktop_config.json)to your Claude Desktop config file:

• Linux: ~/.config/Claude/claude_desktop_config.json
• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

3. Update paths in the config to match your system

4. Restart Claude Desktop

🛠️ Available Tools

Core Operations

• list_ollama_models - Show all available Ollama models
• run_ollama_prompt - Execute prompts with any model (sync/async)
• get_job_status - Check job completion status
• list_jobs - View all running and completed jobs
• cancel_job - Stop running jobs

Script Management

• save_script - Create reusable prompt templates
• list_scripts - View saved templates
• get_script - Read template content
• run_script - Execute templates with variables

Fast-Agent Workflows

• create_fastagent_script - Single-agent scripts
• create_fastagent_workflow - Multi-agent workflows
• run_fastagent_script - Execute agent workflows
• list_fastagent_scripts - View available workflows

System Integration

• run_bash_command - Execute system commands safely
• run_workflow - Multi-step workflow execution

📖 Built-in Prompts

Interactive prompts to guide common tasks:

• ollama_guide - Interactive user guide
• ollama_run_prompt - Simple prompt execution
• model_comparison - Compare multiple models
• fast_agent_workflow - Multi-agent workflows
• script_executor - Template execution
• batch_processing - Multiple prompt processing
• iterative_refinement - Content improvement workflows

📁 Directory Structure

ollama-mcp-server/
├── src/ollama_mcp_server/
│   └── server.py                 # Main server code
├── outputs/                      # Generated output files
├── scripts/                      # Saved script templates
├── workflows/                    # Workflow definitions
├── fast-agent-scripts/          # Fast-agent Python scripts
├── prompts/                      # Usage guides
│   ├── tool_usage_guide.md
│   ├── prompt_templates_guide.md
│   └── setup_guide.md
├── example_mcp_config.json      # Claude Desktop config
└── README.md

🔧 Development

Run Development Server

cd ollama-mcp-server
uv run python -m ollama_mcp_server.server

Debug with MCP Inspector

mcp dev src/ollama_mcp_server/server.py

🛡️ Process Management

The server includes comprehensive process leak prevention:

• Signal Handling: Proper SIGTERM/SIGINT handling
• Background Task Tracking: All async tasks monitored
• Resource Cleanup: Automatic process termination
• Memory Management: Prevents accumulation of zombie processes

Monitor health with:

ps aux | grep mcp | wc -l  # Should show <10 processes

📊 Usage Examples

Simple Prompt Execution

1. Use "ollama_run_prompt" prompt in Claude
2. Specify model and prompt text
3. Get immediate results

Multi-Agent Workflow

1. Use "fast_agent_workflow" prompt
2. Choose workflow type (chain/parallel/router/evaluator)
3. Define agents and initial prompt
4. Monitor execution

Script Templates

1. Create template with save_script
2. Use variables: {variable_name}
3. Execute with run_script
4. Pass JSON variables object

🚨 Troubleshooting

Model not found: Use list_ollama_models for exact names Connection issues: Start Ollama with ollama serve High process count: Server now prevents leaks automatically Job stuck: Use cancel_job to stop problematic tasks

🤝 Contributing

1. Follow the MCP Python SDK development guidelines
2. Use proper type hints and docstrings
3. Test all new features thoroughly
4. Ensure process cleanup in all code paths

📄 License

This project follows the same license terms as the MCP Python SDK.

🙏 Acknowledgments

Built on the Model Context Protocol and Ollama with process management patterns from MCP best practices.

---

Ready to get started? Check the prompts/setup_guide.md for detailed installation instructions!