sandbox-fusion-mcp

# Sandbox Fusion MCP Server This is an implementation of an MCP (Machine Code Protocol) server that interfaces with Sandbox Fusion, providing code interpreter functionality for large language models. ## Features - Supports code execution in multiple programming languages - Supports execution of Jupyter notebooks - Communicates with clients via stdio interaction - Provides resource and tool APIs for large language models to call ## Supported Programming Languages - Python - JavaScript/TypeScript - Bash - R - Julia - Go - Java - C++ - Rust - PHP ## Installation ### Environment Requirements - Python 3.10+ - Conda (recommended for environment management) ### Installation Steps 1. Clone the repository ```bash git clone https://github.com/happyZYM/sandbox-fusion-mcp.git cd sandbox-fusion-mcp ``` 2. Create and activate the Conda environment ```bash conda env create -f environment.yml conda activate mcp-sandbox-fusion ``` ## Configuration The server is configured via environment variables: - `SANDBOX_FUSION_ENDPOINT`: The endpoint address for the Sandbox Fusion API, default is `http://localhost:8000` ## Running Run the server using the provided launch script: ```bash ./launch.sh ``` Or run the Python script directly: ```bash python server.py ``` ## API Description ### Resources The server provides the following resources: 1. `sandbox://languages` - Get the list of supported programming languages 2. `sandbox://endpoint` - Get the current configured Sandbox Fusion endpoint information ### Tools The server provides the following tools: 1. `execute_code` - Execute a code snippet - Parameters: - `code` (string): The code to execute - `language` (string, optional): The programming language, default is "python" - `timeout` (integer, optional): Execution timeout (seconds) 2. `execute_jupyter` - Execute a Jupyter notebook - Parameters: - `notebook_content` (string): Jupyter notebook content (in JSON format) - `timeout` (integer, optional): Execution timeout (seconds) ### Prompt Templates The server provides the following prompt templates: 1. `code_execution_prompt` - Code execution prompt template - Parameters: - `language` (string, optional): The programming language, default is "python" - `description` (string, optional): Description of the code task 2. `jupyter_execution_prompt` - Jupyter notebook execution prompt template - Parameters: - `description` (string, optional): Description of the notebook task ## Client Integration The MCP server communicates with clients via stdio interaction, and clients can communicate using JSON-RPC. Example: List available tools ```json {"jsonrpc": "2.0", "method": "list_tools", "params": {}, "id": 1} ``` ## Troubleshooting If you encounter issues, please check the following log files: - `mcp_server.log`: Main server log - `mcp_stderr.log`: Standard error output log ## Development ### Project Structure ``` sandbox-fusion-mcp/ ├── server.py - MCP server implementation ├── launch.sh - Launch script ├── environment.yml - Conda environment configuration └── README.md - Documentation ``` ### Adding New Features To add new tools or resources, modify `server.py` following this pattern: ```python @mcp_server.tool() def your_tool_name(param1: str, param2: int = 0) -> str: """ Tool description Parameters: - param1: Description of parameter 1 - param2: Description of parameter 2 """ # Implementation logic return json.dumps(result) @mcp_server.resource("sandbox://your_resource") def your_resource_function() -> str: """Resource description""" # Implementation logic return json.dumps(result) ``` ## Notes - Ensure that `stderr` is redirected to the log file to avoid interfering with stdio communication - Tool functions should return results in JSON string format - Handle all possible exceptions and return useful error messages when errors occur ## License MIT