dbt CLI MCP Server

# MCP Server Registration Guide ## Overview This guide explains how to register your Model Context Protocol (MCP) server with Claude or RooCode. Registration allows the AI assistant to discover and use your MCP server's capabilities. ## Configuration for Claude/RooCode To register your MCP server, you need to add it to the appropriate MCP settings file based on the client you're using. ### For RooCode (VSCode Extension) Edit the file at: `/Users/username/Library/Application Support/Code/User/globalStorage/rooveterinaryinc.roo-cline/settings/cline_mcp_settings.json` Add your MCP server configuration: ```json { "mcpServers": { "my-server": { "command": "uv", "args": ["run", "-s", "/path/to/my_mcp_server.py"], "env": { "API_KEY": "your_api_key_here" }, "disabled": false, "alwaysAllow": [] } } } ``` ### For Claude Desktop Edit the file at: `~/Library/Application Support/Claude/claude_desktop_config.json` The format is the same as for RooCode: ```json { "mcpServers": { "my-server": { "command": "uv", "args": ["run", "-s", "/path/to/my_mcp_server.py"], "env": { "API_KEY": "your_api_key_here" }, "disabled": false, "alwaysAllow": [] } } } ``` ## Configuration Fields The MCP server configuration includes the following fields: | Field | Description | Required | |-------|-------------|----------| | `command` | The command to run (typically `"uv"`) | Yes | | `args` | Array of command arguments, including the script path | Yes | | `env` | Object containing environment variables | No | | `disabled` | Boolean indicating if the server is disabled | No (defaults to `false`) | | `alwaysAllow` | Array of function names that don't require permission | No (defaults to `[]`) | ### Command and Args For MCP servers using uv and inline dependencies: ```json "command": "uv", "args": ["run", "-s", "/absolute/path/to/my_mcp_server.py"] ``` For MCP servers with environment files: ```json "command": "uv", "args": ["run", "--env-file", "/path/to/.env", "-s", "/path/to/my_mcp_server.py"] ``` ### Environment Variables The `env` field contains environment variables as key-value pairs: ```json "env": { "API_KEY": "your_api_key_here", "DEBUG": "true", "TIMEOUT_SECONDS": "30" } ``` ### Disabled Flag The `disabled` flag determines whether the MCP server is active: ```json "disabled": false // Server is active "disabled": true // Server is inactive ``` ### Always Allow The `alwaysAllow` array lists functions that can be called without explicit user permission: ```json "alwaysAllow": ["get_weather", "convert_units"] ``` Use this cautiously, as it bypasses the normal permission system. ## Environment Variables Management When configuring your MCP server, follow these rules for managing environment variables: 1. **NEVER hardcode sensitive keys** in your MCP server code 2. **ALWAYS use environment variables** for all API keys and secrets 3. **ALWAYS provide clear error messages** when required environment variables are missing 4. **ALWAYS document all required environment variables** in your README 5. **ALWAYS include a `.env.example` file** in your project showing required variables (without values) Example environment variable validation in your code: ```python api_key = os.environ.get("API_KEY") if not api_key: raise ValueError("API_KEY environment variable is required") ``` ## Registration Best Practices 1. **Use descriptive server names**: Choose a name that clearly indicates the server's purpose 2. **Use absolute paths**: Always use absolute paths to avoid working directory issues 3. **Include version information**: Consider including version in the server name or documentation 4. **Document environment variables**: Clearly document all required environment variables 5. **Test before registration**: Verify the server works with the `--test` flag before registering 6. **Security checks**: Ensure sensitive information is only in environment variables, not hardcoded ## Multi-Environment Configuration For running the same MCP server in different environments: ```json { "mcpServers": { "weather-prod": { "command": "uv", "args": ["run", "-s", "/path/to/weather_mcp.py"], "env": { "API_KEY": "production_key_here", "ENVIRONMENT": "production" }, "disabled": false }, "weather-dev": { "command": "uv", "args": ["run", "-s", "/path/to/weather_mcp.py"], "env": { "API_KEY": "development_key_here", "ENVIRONMENT": "development", "DEBUG": "true" }, "disabled": false } } } ``` ## Security Considerations When registering MCP servers, consider these security practices: 1. **API Keys**: Store API keys only in the environment variables, never in code 2. **Least Privilege**: Use API keys with the minimum necessary permissions 3. **Rotate Credentials**: Regularly rotate API keys and other credentials 4. **Validation**: Always validate and sanitize inputs to prevent injection attacks 5. **Authorization**: Implement authorization checks for sensitive operations 6. **Rate Limiting**: Implement rate limiting to prevent abuse ## Troubleshooting Registration If you encounter issues with registration: 1. **Verify path**: Ensure the path to your script is correct and absolute 2. **Check permissions**: Verify the script has execute permissions 3. **Test directly**: Run with `uv run -s my_mcp_server.py --test` to verify it works 4. **Check environment**: Ensure all required environment variables are set 5. **Review logs**: Look for error messages in the console or logs 6. **Check JSON syntax**: Ensure your config file has valid JSON syntax ## Next Steps After registering your MCP server: - Restart the Claude application or RooCode extension to load the new configuration - Test the server by asking Claude to use one of your server's tools - Monitor logs for any errors or issues