Repoburg
Login

Integrations: Model Context Protocol (MCP)

The Model Context Protocol (MCP) is a powerful feature that allows Repoburg to integrate deeply with external tools running on your local machine. It provides a standardized way for the AI to discover and execute functions on these tools, treating them as extensions of its own capabilities.

This is a more advanced integration method than the direct API/WebSocket approach used by the AI Studio userscript, as it enables a true, bidirectional, and tool-aware workflow.

How It Works

The MCP workflow follows a clear sequence:

  1. Configuration: You define how to connect to an external tool server on the MCP Server Management page. This is typically a command that starts the server process.
  2. Connection & Discovery: When you activate a server configuration, Repoburg's backend starts the server process and connects to it. It then asks the server for a list of available tools, including their names, descriptions, and expected input schemas.
  3. Dynamic Prompt Injection: The discovered tools are formatted and dynamically injected into the System Prompt wherever you place the {{MCP_SERVERS}} placeholder. This makes the AI aware of the new tools it can use.
  4. AI Invocation: When the AI determines that one of the external tools is suitable for a task, it will generate a special use_mcp_tool action in its plan.
  5. Execution: Repoburg's backend receives this action and executes the specified tool on the corresponding MCP server with the arguments provided by the AI.
  6. Feedback Loop: The output from the tool is then sent back to the AI in the next prompt, allowing it to analyze the results and continue with its task.

The MCP Management Page

You can manage all your MCP connections from the MCP Servers page, accessible from the home dashboard.

Key Features

  • Add New Server: You can add a new server by providing a JSON configuration that specifies a unique name for the server and the command needed to run it.
    { "my-ide-server": { "command": "path/to/ide/mcp-server-executable", "args": ["--port", "8000"], "env": {"API_KEY": "secret"} } }
  • Activate/Deactivate Servers: You can toggle servers on or off. Activating a server establishes the connection and discovers its tools. Deactivating it terminates the connection.
  • View Tools: For each active server, you can see a list of all tools it provides. You can expand each tool to view its description and the expected input parameters (its schema).
  • Activate/Deactivate Individual Tools: You have fine-grained control to enable or disable specific tools. Disabled tools will not be included in the system prompt and cannot be used by the AI. This is useful for preventing the AI from using unstable or irrelevant tools.
  • Batch Actions: You can select multiple tools and activate or deactivate them all at once.

Use Cases

MCP opens up a vast range of possibilities. Any application that can expose its functionality via the simple MCP server protocol can be integrated.

  • IDE Integration: Connect to a plugin in your JetBrains IDE or VS Code to get a list of run configurations, trigger a build, or run tests.
  • Database Tools: Allow the AI to query a database schema or run SQL queries.
  • Linters & Formatters: Expose code linting or formatting tools directly to the AI.