mcpman 0.2.3

Model Context Protocol (MCP) Manager - a tool for managing MCP servers

MCPMan (MCP Manager)

MCPMan orchestrates interactions between LLMs and Model Context Protocol (MCP) servers, making it easy to create powerful agentic workflows.

Quick Start

Run MCPMan instantly without installing using uvx:

# Use the calculator server to perform math operations
uvx mcpman -c server_configs/calculator_server_mcp.json -i openai -m gpt-4.1-mini -p "What is 1567 * 329 and then divide by 58?"

# Use the datetime server to check time in different timezones
uvx mcpman -c server_configs/datetime_server_mcp.json -i gemini -m gemini-2.0-flash-001 -p "What time is it right now in Tokyo, London, and New York?"

# Use the filesystem server with Ollama for file operations
uvx mcpman -c server_configs/filesystem_server_mcp.json -i ollama -m llama3:8b -p "Create a file called example.txt with a sample Python function, then read it back to me"

# Use the filesystem server with LMStudio's local models
uvx mcpman -c server_configs/filesystem_server_mcp.json -i lmstudio -m qwen2.5-7b-instruct-1m -p "Create a simple JSON file with sample data and read it back to me"

You can also use uv run for quick one-off executions directly from GitHub:

uv run github.com/ericflo/mcpman -c server_configs/calculator_server_mcp.json -i openai -m gpt-4.1-mini -p "What is 256 * 432?"

Core Features

• One-command setup: Manage and launch MCP servers directly
• Tool orchestration: Automatically connect LLMs to any MCP-compatible tool
• Detailed logging: JSON structured logs for every interaction
• Multiple LLM support: Works with OpenAI, Google Gemini, Ollama, LMStudio and more
• Flexible configuration: Supports stdio and SSE server communication

Installation

# Install with pip
pip install mcpman

# Install with uv
uv pip install mcpman

# Install from GitHub
uvx pip install git+https://github.com/ericflo/mcpman.git

Basic Usage

mcpman -c <CONFIG_FILE> -i <IMPLEMENTATION> -m <MODEL> -p "<PROMPT>"

Examples:

# Use local models with Ollama for filesystem operations
mcpman -c ./server_configs/filesystem_server_mcp.json \
       -i ollama \
       -m codellama:13b \
       -p "Create a simple bash script that counts files in the current directory and save it as count.sh"

# Use OpenAI with multi-server config
mcpman -c ./server_configs/multi_server_mcp.json \
       -i openai \
       -m gpt-4.1-mini \
       -s "You are a helpful assistant. Use tools effectively." \
       -p "Calculate 753 * 219 and tell me what time it is in Sydney, Australia"

Server Configuration

MCPMan uses JSON configuration files to define the MCP servers. Examples:

Calculator Server:

{
  "mcpServers": {
    "calculator": {
      "command": "python",
      "args": ["-m", "mcp_servers.calculator"],
      "env": {}
    }
  }
}

DateTime Server:

{
  "mcpServers": {
    "datetime": {
      "command": "python",
      "args": ["-m", "mcp_servers.datetime_utils"],
      "env": {}
    }
  }
}

Filesystem Server:

{
  "mcpServers": {
    "filesystem": {
      "command": "python",
      "args": ["-m", "mcp_servers.filesystem_ops"],
      "env": {}
    }
  }
}

Key Options

Option	Description
-c, --config <PATH>	Path to MCP server config file
-i, --implementation <IMPL>	LLM implementation (openai, gemini, ollama, lmstudio)
-m, --model <MODEL>	Model name (gpt-4.1-mini, gemini-2.0-flash-001, llama3:8b, qwen2.5-7b-instruct-1m, etc.)
-p, --prompt <PROMPT>	User prompt (text or file path)
-s, --system <MESSAGE>	Optional system message
--base-url <URL>	Custom endpoint URL
--temperature <FLOAT>	Sampling temperature (default: 0.7)
--max-tokens <INT>	Maximum response tokens
--no-verify	Disable task verification

API keys are set via environment variables: OPENAI_API_KEY, GEMINI_API_KEY, etc.

Why MCPMan?

• Standardized interaction: Unified interface for diverse tools
• Simplified development: Abstract away LLM-specific tool call formats
• Debugging support: Detailed JSONL logs for every step in the agent process
• Local or cloud: Works with both local and cloud-based LLMs

Currently Supported LLMs

• OpenAI (GPT-4.1, GPT-4.1-mini, GPT-4.1-nano)
• Anthropic Claude (claude-3-7-sonnet-20250219, etc.)
• Google Gemini (gemini-2.0-flash-001, etc.)
• OpenRouter
• Ollama (llama3, codellama, etc.)
• LM Studio (Qwen, Mistral, and other local models)

Development Setup

# Clone and setup
git clone https://github.com/ericflo/mcpman.git
cd mcpman

# Create environment and install deps
uv venv
source .venv/bin/activate  # Linux/macOS
# or .venv\Scripts\activate  # Windows
uv pip install -e ".[dev]"

# Run tests
pytest tests/

Project Structure

• src/mcpman/: Core source code
• mcp_servers/: Example MCP servers for testing
• server_configs/: Example configuration files
• logs/: Auto-generated structured JSONL logs

License

Licensed under the Apache License 2.0.