MCP Development Agent

A development agent based on the Model Context Protocol (MCP) that provides memory management, documentation search, and Git/GitHub integration features, available both as a CLI tool and as an SSE server.

Features

• 🧠 Memory System: Store and retrieve information using embeddings via ChromaDB
• 📚 Documentation Search: Search through documentation with AI assistance
• 🔄 Git Integration: Query commit history and issues
• 🔍 Code Analysis: Static code analysis using tree-sitter
• 🌐 GitHub Integration: Search repositories, issues, pull requests, projects, and code
• 💡 AI Assistant: Local AI-powered features using Ollama
• 💻 CLI Interface: Interactive command-line interface
• 🌐 Server Mode: Server-Sent Events (SSE) support

Requirements

• Python 3.x
• ChromaDB
• GitPython
• FastMCP
• PyGithub
• tree-sitter
• Ollama (https://ollama.ai)

Installation

Global Installation (Recommended)

1. Install Ollama following instructions at https://ollama.ai

2. Pull required models:

ollama pull codellama
ollama pull llama3
ollama pull mistral

3. Clone and install:

For Windows:

git clone [repository-url]
cd mcp-dev-agent
.\install.bat

For Linux/MacOS:

git clone [repository-url]
cd mcp-dev-agent
./install.sh

This will:

• Create a virtual environment
• Install all dependencies
• Configure the global command mcp-dev-agent-cli
• Set up the required directory structure

4. Configure your GitHub token in the .env file

After installation, you can use the mcp-dev-agent-cli command from any git repository!

Manual Installation (Alternative)

1. Install Ollama following instructions at https://ollama.ai

2. Pull required models:

ollama pull codellama
ollama pull llama3
ollama pull mistral

3. Clone the repository:

git clone [repository-url]
cd mcp-dev-agent

4. Install dependencies:

pip install -r requirements.txt

5. Configure environment variables (copy .env.example to .env):

cp .env.example .env

Setting up GitHub Token

1. Go to GitHub Personal Access Tokens
2. Click "Generate new token (classic)"
3. Give it a name (e.g., "MCP Dev Agent")
4. Select these scopes:
   • repo (Full control of repositories)
   • read:org (Read organization info)
5. Click "Generate token"
6. IMPORTANT: Copy your token immediately! You won't see it again.
7. Add it to your .env file:

   GITHUB_TOKEN=ghp_your_actual_token_here

8. Or set it in your environment:

   export GITHUB_TOKEN=ghp_your_actual_token_here

Note: Replace ghp_your_actual_token_here with the actual token you copied from GitHub.

7. Set up code parsers:

python setup_parsers.py

Usage

Global CLI

The MCP Dev Agent can be used from any git repository using the global command:

mcp-dev-agent-cli

This command will:

• Automatically set up required directories in your project
• Use the correct Python environment
• Maintain proper context isolation between projects

Traditional CLI Mode

python main.py --mode cli

Available commands:

Memory

• /memory add <content> - Add a new general memory
• /memory get <query> - Search existing memories
• /memory repo add <content> - Add a repository-specific memory
• /memory repo get <query> - Search repository-specific memories

Git and GitHub

• /git commits [number] - Show commit history (default: last 5)
• /git issues - List local repository issues
• /git info - Show detailed repository information
• /git diff - Show pending changes (staged and unstaged)
• /github repo <owner/repo> - Show repository details
• /github issues <owner/repo> [state] - List issues (state: open/closed)
• /github prs <owner/repo> [state] - List pull requests (state: open/closed)
• /github project <org> <number> - Show project information
• /github summarize <owner/repo> <issue_number> - Generate issue summary using GPT
• /github search <query> [language] - Search code on GitHub

Code Analysis

• /code analyze <file> [language] - Analyze code structure and provides:
  • High-level overview of the code's purpose
  • Key components and their responsibilities
  • Notable patterns or techniques used
  • Potential improvements and best practices
  • Security considerations when relevant
  • Structural analysis (functions, classes, imports)

Example:

mcp-dev-agent-cli
> /code analyze src/main.py

Documentation

• /docs <query> - Search documentation
• exit - Exit CLI mode

Advanced Features

Code Analysis

The system uses tree-sitter to analyze code in:

• Python
• JavaScript
• TypeScript

Analysis includes:

• Function and method extraction
• Class identification
• Import mapping
• Memory integration for context

GitHub Integration

• Detailed repository information
• Issue search and listing
• Pull request management
• GitHub Projects integration
• Code search with language filtering
• Enriched metadata
• Automatic issue summarization with GPT

Enhanced Memory System

• Support for different context types
• Enriched metadata
• Contextual search
• Temporal history
• Code analysis integration
• Automatic summary storage
• Git state contextualization

Server Mode (SSE)

python main.py
# or
python main.py --mode server

The SSE server enables integration with other applications through the MCP protocol.

Project Structure

.
├── main.py              # Application entry point
├── setup_parsers.py     # Code parser configuration
├── tools/
│   ├── memory_tool.py   # Memory management via ChromaDB
│   ├── doc_tool.py      # Documentation search
│   ├── git_tool.py      # Git integration
│   └── github_tool.py   # GitHub integration and code analysis
└── docs/
    └── api_reference.md # API reference documentation

AI Model Configuration

The agent uses Ollama's local models for AI features. The following models are configured:

1. CodeLlama (default) - Optimized for code-related tasks
2. Llama 2 - General purpose tasks
3. Mistral - Alternative model for various tasks

The .agent.json file is already configured with:

• CodeLlama as default model
• Response caching (24 hours)
• Interaction logging

Model Parameters

You can adjust model parameters by editing the .agent.json file:

• temperature: Controls response creativity (0.0 to 1.0)
• context_window: Maximum context length
• top_p: Controls response diversity

Using Different Models

Different features use specialized models:

• Code documentation: CodeLlama
• Issue summarization: CodeLlama
• General queries: Llama 2

No API keys required - all AI features run locally through Ollama!