Mindpilot MCP

What abrinsmead/mindpilot-mcp MCP server does

abrinsmead/mindpilot-mcp MCP server connects an AI coding agent to a local Mermaid diagram viewer. The agent can turn requests about source code, architecture, or processes into Mermaid syntax, which the server renders in a browser-based interface. Example uses include mapping WebSocket connection states, creating a C4 context diagram, and illustrating an OAuth sequence.

The project is intended for interactive analysis rather than remote diagram hosting. Its local processing model keeps diagrams on the local machine instead of sending them to a separate diagram service. The diagrams can also be exported as vector images.

How it works

The MCP process receives Mermaid syntax generated by the agent and serves the rendered diagrams through a local web application. The default interface is available at http://localhost:4000. The server saves diagram data under ~/.mindpilot/data/ unless a different location is provided.

Multiple MCP clients can use the same running instance. The first client starts the shared web server; later clients connect to it and use the same history and browser interface. After the final MCP client disconnects, the server shuts down after approximately one minute.

Setup and configuration

abrinsmead/mindpilot-mcp MCP server requires Node.js 20 or newer. Install it as a local stdio MCP process with the @mindpilot/mcp npm package:

npx -y @mindpilot/mcp@latest

Claude Code, Cursor, Windsurf, and other MCP-capable clients can launch the process with npx and the package name. Configuration examples in the project also cover VS Code and Zed. The exact configuration format depends on the host client.

Use --port to change the web server port, such as --port 5555 when port 4000 is already occupied. Use --data-path to choose another directory for saved diagrams. The --disable-analytics flag turns off the anonymous usage tracking described by the project.

Tools and capabilities

The server supports agent-directed diagram generation for subjects such as:

• Code structure and legacy code
• System and software architecture
• State machines
• Process flows
• OAuth flows
• C4 context diagrams

The resulting Mermaid diagrams are displayed in the local web interface and can be exported as vector images. Persistent diagram history is shared among concurrently connected MCP clients.

Limitations and notes

The server depends on a local Node.js runtime and browser access to its local web application. It does not describe a hosted endpoint or a cloud storage service. Diagram quality depends on the Mermaid syntax produced by the connected language model; the MCP server is described as accepting and rendering that syntax rather than generating the underlying analysis itself.

Anonymous usage data is collected by default according to the README. Add --disable-analytics to the MCP arguments if that collection is not desired. A custom data path can support local synchronization workflows, but the project does not document a built-in synchronization provider.

The README also points to a separate Mermaid viewer agent skill that does not require a local MCP server. That alternative is distinct from abrinsmead/mindpilot-mcp MCP server.