mage-mcp-server

# Mage.ai MCP Integration A comprehensive Model Context Protocol (MCP) integration for [Mage.ai](https://mage.ai/), allowing AI assistants like Anthropic's Claude to efficiently work with Mage data pipelines. ## Features - **Browse Pipelines**: Explore all Mage.ai pipelines and their components - **View and Edit Code**: Read and modify block code - **Execute Pipelines**: Run entire pipelines or individual blocks - **Search Functionality**: Find blocks by name, type, or content - **Pipeline Creation**: Create new pipelines and blocks from scratch ## Installation ```bash # Clone the repository git clone https://github.com/yourusername/mage-mcp.git cd mage-mcp # Install dependencies pip install -r requirements.txt ``` ## Configuration The server can be configured using environment variables or command-line arguments: - `MAGE_API_URL`: URL for the Mage.ai API (default: `http://localhost:6789/api/`) - `MAGE_API_KEY`: API key for authentication (if required) Create a `.env` file for persistent configuration: ``` MAGE_API_URL=http://localhost:6789/api/ MAGE_API_KEY=your_api_key_if_needed ``` ## Usage ### Running the Server Use the provided script to start the MCP server: ```bash # Using stdio transport (default) python run_server.py # Using SSE transport python run_server.py --transport sse --host localhost --port 3000 # Additional options python run_server.py --api-url http://your-mage-instance:6789/api/ --api-key your_key --debug ``` ### Installing with Claude Desktop For the best experience, install this server directly with Claude Desktop: ```bash # Using mcp CLI mcp install run_server.py --name "Mage.ai Manager" # With environment variables mcp install run_server.py -v MAGE_API_URL=http://localhost:6789/api/ -v MAGE_API_KEY=your_key ``` ### Development Mode For testing and development: ```bash mcp dev run_server.py ``` ## MCP Resources and Tools ### Resources - `mage://pipelines` - List all pipelines - `mage://pipelines/{pipeline_uuid}` - Get pipeline details - `mage://pipelines/{pipeline_uuid}/blocks` - List blocks in a pipeline - `mage://pipelines/{pipeline_uuid}/blocks/{block_uuid}` - Get block details - `mage://pipelines/{pipeline_uuid}/blocks/{block_uuid}/content` - Get block code - `mage://pipelines/{pipeline_uuid}/content` - Get entire pipeline content - `mage://pipelines/{pipeline_uuid}/execution` - Get pipeline execution details - `mage://blocks/search/{query}` - Search blocks by name, type or content ### Tools - `list_all_pipelines()` - List all pipelines with details - `get_pipeline_details(pipeline_uuid)` - Get pipeline information - `list_pipeline_blocks(pipeline_uuid)` - List blocks in a pipeline - `get_block_content(pipeline_uuid, block_uuid)` - Get block code - `create_pipeline(name, description, pipeline_type)` - Create a new pipeline - `create_block(pipeline_uuid, name, block_type, language, content, upstream_blocks)` - Create a block - `update_block_content(pipeline_uuid, block_uuid, content)` - Update block code - `execute_pipeline(pipeline_uuid)` - Run a pipeline - `execute_block(pipeline_uuid, block_uuid)` - Run a specific block - `get_pipeline_code(pipeline_uuid)` - Get complete pipeline code ## Example Interactions ``` User: Show me all the pipelines in Mage.ai. Claude: Let me check what pipelines are available in your Mage.ai instance. [Lists all pipelines with descriptions and types] User: Can you show me the blocks in the data_ingestion pipeline? Claude: Here are all the blocks in the data_ingestion pipeline: [Lists all blocks with their types, languages, and status] User: I need to modify the 'transform_data' block to fix a bug. Claude: Let me help you with that. First, let's look at the current code: [Shows the current block code] Here's how we can fix the bug: [Explains the issue and suggests changes] Would you like me to update the block with these changes? User: Yes, please update it. Claude: I've updated the block. Here's the new content: [Shows the updated code] The changes have been saved successfully. ``` ## Contributing Contributions are welcome! Please feel free to submit a Pull Request. ## License This project is licensed under the MIT License - see the LICENSE file for details.