MCP-CodeSavant

CodeSavant is a WIP project.

CodeSavant is a Model Context Protocol (MCP) server that provides code manipulation, execution, and version control capabilities. It allows AI assistants to read, write, and execute code while maintaining a history of changes.

Features

• Read and write code files with line-specific operations
• Execute code in multiple programming languages (Python, Node.js)
• Execute shell commands in a controlled environment
• Track and manage code changes with version control
• Search within code files
• Revert to previous versions of code

Installation

1. Clone the repository:

git clone https://github.com/twolven/mcp-codesavant.git
cd mcp-codesavant

2. Install the required dependencies:

pip install -r requirements.txt

3. Add the server configuration to your Claude Desktop config.json:

{
  "mcpServers": {
    "codesavant": {
      "command": "python",
      "args": ["path/to/codesavant.py"]
    }
  }
}

Directory Structure

The server creates and manages the following directory structure:

workspaces/
├── project1/
│   ├── .code_history.json
│   └── [code files]
├── project2/
│   ├── .code_history.json
│   └── [code files]
└── ...

Tool Reference

Detailed Usage

1. read_code_file

Read contents of a code file, optionally searching for specific sections.

{
    "project": "string",     // Project name
    "path": "string",        // Path to file relative to project workspace
    "search": "string"       // (Optional) Text/pattern to search for in file
}

Response:

{
    "success": true,
    "timestamp": 1234567890,
    "data": {
        "content": "string",      // File content
        "start_line": number,     // (Only if search used) Starting line of found section
        "end_line": number       // (Only if search used) Ending line of found section
    }
}

2. write_code_file

Write or update specific lines in a code file.

{
    "project": "string",     // Project name
    "path": "string",        // Path to file relative to workspace
    "content": "string",     // Content to write (just the lines being changed)
    "start_line": number,    // Starting line number for the change
    "end_line": number      // (Optional) Ending line number for the change
}

Response:

{
    "success": true,
    "timestamp": 1234567890,
    "data": {
        "diff": {
            "changes": [           // List of changes made
                [string, number, number, number, number]  // (type, old_start, old_end, new_start, new_end)
            ],
            "timestamp": number    // When the change was made
        }
    }
}

3. get_code_history

Get change history for a code file.

{
    "path": "string"        // Path to file relative to workspace
}

Response:

{
    "success": true,
    "timestamp": 1234567890,
    "data": {
        "history": [
            {
                "changes": [       // List of changes made
                    [string, number, number, number, number]
                ],
                "timestamp": number
            }
        ]
    }
}

4. execute_code_command

Execute a code-related shell command.

{
    "command": "string",     // Shell command to execute
    "timeout": number       // (Optional) Command timeout in seconds (default: 30)
}

Response:

{
    "success": true,
    "timestamp": 1234567890,
    "data": {
        "state": "success|error|timeout|cancelled",
        "output": "string",        // Command output
        "error": "string",         // Error message if any
        "runtime": number,         // Execution time in seconds
        "exit_code": number       // Command exit code
    }
}

5. execute_code

Execute code in specified language.

{
    "code": "string",        // Code to execute
    "language": "string",    // Programming language ("python" or "node")
    "timeout": number       // (Optional) Execution timeout in seconds (default: 30)
}

Response:

{
    "success": true,
    "timestamp": 1234567890,
    "data": {
        "state": "success|error|timeout|cancelled",
        "output": "string",        // Code execution output
        "error": "string",         // Error message if any
        "runtime": number,         // Execution time in seconds
        "exit_code": number       // Execution exit code
    }
}

6. revert_to_version

Revert a code file to a specific version.

{
    "path": "string",        // Path to file relative to workspace
    "timestamp": number     // Timestamp of version to revert to
}

Response:

{
    "success": true,
    "timestamp": 1234567890,
    "data": {
        "diff": {
            "changes": [           // List of changes made
                [string, number, number, number, number]
            ],
            "timestamp": number    // When the reversion was made
        }
    }
}

7. read_code_file_lines

Read specific lines from a code file.

{
    "project": "string",     // Project name
    "path": "string",        // Path to file relative to project workspace
    "start_line": number,    // Starting line number to read
    "end_line": number      // (Optional) Ending line number to read
}

Response:

{
    "success": true,
    "timestamp": 1234567890,
    "data": {
        "content": "string"        // Content of the specified lines
    }
}

Error Handling

The server provides detailed error responses in the following format:

{
    "success": false,
    "timestamp": 1234567890,
    "data": null,
    "error": "Error message"
}

Error types include:

• CodeFileError: File operation errors
• CodeValidationError: Code validation errors
• CodeExecutionError: Code execution errors

Language Support

Currently supported languages for code execution:

• Python (using system Python interpreter)
• Node.js (using node command)

Each language execution creates a temporary file in the workspace directory and executes it with appropriate interpreter.

Contributing

1. Fork the repository
2. Create your feature branch
3. Commit your changes
4. Push to the branch
5. Create a new Pull Request

License

This project is licensed under the MIT License - see the LICENSE file for details.

Author

Todd Wolven - (https://github.com/twolven)

Acknowledgments

• Built with the Model Context Protocol (MCP) by Anthropic
• Developed for use with Anthropic’s Claude