What is Agent Mcp Army

agent-mcp-army is a powerful multi-agent system built with Archon, designed to orchestrate specialized AI agents that perform various tasks through third-party services like Slack, Firecrawl, and Airtable. It leverages the Model Context Protocol (MCP) and Pydantic AI to enhance the capabilities of AI agents by distributing tasks among subagents.

Use cases

Use cases for agent-mcp-army include managing Airtable databases, performing web searches with the Brave Search Agent, handling file operations with the Filesystem Agent, and automating Slack communications. It is ideal for organizations looking to streamline workflows across different platforms.

How to use

To use agent-mcp-army, clone the repository, set up a virtual environment, install the required Python dependencies, and configure your API keys in the .env file. After setup, you can interact with the orchestration agent to delegate tasks to specialized subagents.

Key features

Key features of agent-mcp-army include a primary orchestration agent that delegates tasks, specialized subagents for various services (e.g., Airtable, GitHub, Slack), and the ability to handle multiple tools efficiently without overwhelming a single agent.

Where to use

agent-mcp-army can be used in various fields including project management, data extraction, communication automation, and web searching, making it suitable for teams that rely on multiple third-party services.

MCP Agent Army

A powerful multi-agent system built with Archon, the AI agent that builds other AI Agents. This system leverages the Model Context Protocol (MCP) and Pydantic AI to orchestrate specialized AI agents to perform various tasks through third-party services (Slack, Firecrawl, Airtable, etc.).

Specialized agents are important because LLMs get overwhelmed very easily if you give a single agent too many tools.
Splitting the tools for each service into subagents is the best way to give an agent system many capabilities while
still keeping each individual prompt relatively short.

See prompt.txt for the initial prompt I gave to Archon to build this AI agent army. Keep in mind I had to iterate on the agent for a few prompts with Archon to get it right, but it still gave a fantastic starting point after one shot!

Overview

This system uses a primary orchestration agent that delegates tasks to specialized subagents, each with expertise in a specific third-party service:

• Airtable Agent: Manages Airtable databases and records
• Brave Search Agent: Performs web searches and retrieves information
• Filesystem Agent: Handles file operations and directory management
• GitHub Agent: Interacts with GitHub repositories, issues, and PRs
• Slack Agent: Sends messages and manages Slack communications
• Firecrawl Agent: Extracts data from websites through web crawling

Requirements

• Python 3.9+
• Node.js and npm (for MCP servers)
• API keys for various services (see .env.example)

Installation

1. Clone this repository
2. Set up a virtual environment:

   # On Windows
   python -m venv venv
   venv\Scripts\activate
   
   # On macOS/Linux
   python3 -m venv venv
   source venv/bin/activate
   

3. Install Python dependencies:

   pip install -r requirements.txt
   

4. Copy .env.example to .env and fill in your API keys and configuration

Environment Variables

Set the following environment variables in your .env file:

• PROVIDER: Your LLM provider (OpenAI, OpenRouter, Ollama)
• BASE_URL: API base URL for your LLM provider
• LLM_API_KEY: Your LLM API key
• MODEL_CHOICE: The model to use (e.g., gpt-4o-mini)
• BRAVE_API_KEY: API key for Brave Search
• AIRTABLE_API_KEY: API key for Airtable
• GITHUB_TOKEN: Personal access token for GitHub
• SLACK_BOT_TOKEN: Bot token for Slack
• SLACK_APP_TOKEN: App token for Slack
• FIRECRAWL_API_KEY: API key for Firecrawl

Usage

Run the main script:

python mcp_agent_army.py

Enter your requests at the prompt. The primary agent will analyze your request and delegate it to the appropriate specialized agent.

Example requests:

• “Search for the latest AI research papers on multi-agent systems”
• “Create a new file called test.txt with ‘Hello World’ content”
• “Check the status of my GitHub repository issues”
• “Send a message to the #general channel in Slack”
• “Extract product information from the Amazon page for iPhone 15”
• “Create a new record in my Airtable database”

Architecture

The system uses AsyncExitStack to manage all MCP servers in a single context, making it efficient and robust. Each subagent is initialized with its own MCP server and system prompt that defines its expertise.

The primary agent has tools to invoke each subagent, allowing it to delegate tasks based on the user’s request.

License

MIT