Modelbrain MCP

What ModelBrain/modelbrain-mcp does

ModelBrain/modelbrain-mcp connects an MCP-compatible AI client to the ModelBrain desktop application. ModelBrain provides a local memory layer for AI assistants, with data handled through software running on the user’s own computer. The package is a wrapper and distribution target; it does not contain or install the ModelBrain engine.

The ModelBrain/modelbrain-mcp MCP server is intended for developers who want to add ModelBrain’s memory functions to an assistant that supports MCP over standard input and output. It is relevant when local storage and a separately installed desktop application fit the deployment requirements.

How it works

When started, the package searches the current machine for ModelBrain’s installed mcp_server binary. That binary communicates with the local ModelBrain daemon and speaks MCP over stdio. The wrapper therefore depends on the ModelBrain desktop app and its local components instead of implementing memory storage itself.

If ModelBrain is absent, the package reports where it can be obtained and exits. It does not silently continue without the required application. The ModelBrain/modelbrain-mcp MCP server supports the operating systems identified by the project: macOS on Apple Silicon and Windows on x64 or arm64 hardware.

Setup and configuration

Install ModelBrain separately before registering this package. The project documentation points to the ModelBrain download page and states that the desktop app must already be installed.

For clients that launch MCP servers with npx, add an entry similar to the following to the client’s MCP configuration:

{
  "mcpServers": {
    "modelbrain": {
      "command": "npx",
      "args": ["-y", "modelbrain-mcp"]
    }
  }
}

Claude Desktop, Claude Code, and Cursor users generally do not need this wrapper configuration because the ModelBrain app registers itself with those clients directly. Other clients can use the stdio configuration when they support this launch method.

Tools and capabilities

The available ModelBrain tool surface includes:

• remember for storing information in ModelBrain.
• recall for retrieving stored information.
• forget for removing information.
• expand for expanding or developing memory context.
• connector_sync for synchronizing through ModelBrain connectors.

These tools are provided by the installed ModelBrain application and local daemon. The wrapper makes them available through MCP but does not describe a separate remote service or hosted data store.

Limitations and notes

The ModelBrain/modelbrain-mcp MCP server requires ModelBrain to be installed and running locally, and it supports only macOS Apple Silicon and Windows x64 or arm64 according to the supplied requirements. Installing the npm package alone does not provide the memory engine.

The wrapper package is licensed under MIT, while ModelBrain itself is separate proprietary software with its own terms. Review ModelBrain’s documentation for its complete tool behavior and client connection guides. For security details, consult the repository’s SECURITY.md file.