Excalidraw Architect Mcp

What is Excalidraw Architect Mcp?

An MCP server that lets AI assistants generate, edit, and convert architectural diagrams as Excalidraw files. It separates diagram structure from layout: the AI defines components and connections, while the server uses a graph layout algorithm to produce clean, styled diagrams. Designed for developers onboarding codebases, system designers, and anyone documenting architecture.

How to use Excalidraw Architect Mcp?

Install via pip install excalidraw-architect-mcp or run with uvx excalidraw-architect-mcp. Configure it as an MCP server in Cursor, Windsurf, or other IDEs using stdio transport. Then ask your AI to create, modify, or convert diagrams using natural language. An optional Diagram Design Skill file improves results.

Key features of Excalidraw Architect Mcp

• Auto-layout using the Sugiyama hierarchical algorithm
• 50+ technology mappings with automatic styling (Kafka, PostgreSQL, etc.)
• Interactive editing of existing diagrams via natural language
• Hub node visualization (gateways, load balancers stretch across services)
• Mermaid flowchart conversion to Excalidraw
• Fully offline – no API keys required

Use cases of Excalidraw Architect Mcp

• Quickly generate architecture diagrams when onboarding a new codebase
• Iterate on system design during brainstorming sessions
• Keep documentation up to date by modifying diagrams with natural language

FAQ from Excalidraw Architect Mcp

How is it different from Mermaid or plain Excalidraw?

Mermaid diagrams are quick but lack manual repositioning; plain Excalidraw requires LLMs to hallucinate coordinates, causing overlapping boxes. Excalidraw Architect Mcp lets the AI focus on structure while the server handles layout and styling automatically.

What are the runtime requirements?

Python 3.x and the excalidraw-architect-mcp package. It can also run via uv without installation. No network access or API keys are needed.

Where are the diagrams stored?

Diagrams are saved as .excalidraw files on your local machine. You can open them with the Excalidraw VS Code extension or at excalidraw.com.

Are there any limits on diagram size?

Yes. For best readability, aim for 6–15 nodes in architecture diagrams and 10–25 nodes in detailed flows. Very large diagrams should be split into multiple focused diagrams.

What transport and authentication does it use?

It uses stdio transport for MCP communication. No authentication or external service calls are required.