@drawio/mcp

Enables LLMs to open diagram files (draw.io XML, Mermaid, CSV, SVG) directly in the draw.io web editor via MCP protocol, establishing a bidirectional communication channel between the LLM and the editor. Uses MCP resource URIs to reference local or remote diagram files and translates them into draw.io-compatible formats, allowing the LLM to initiate editor sessions with pre-loaded diagrams for visualization and interactive editing.

Converts Mermaid diagram syntax (flowcharts, sequence diagrams, class diagrams, etc.) into draw.io XML format for rendering and editing in the draw.io editor. The conversion process parses Mermaid syntax, maps diagram elements to draw.io shape primitives, and generates valid XML with positioning, styling, and connector information, allowing LLMs to author diagrams in Mermaid and visualize them in draw.io's interactive editor.

Transforms CSV data into draw.io table diagrams with structured rows, columns, and styling. The conversion parses CSV headers and rows, creates draw.io table primitives with cell formatting, and generates a visual representation suitable for data modeling, entity-relationship diagrams, or data flow documentation. Enables LLMs to convert tabular data into visual diagram format for inclusion in draw.io projects.

Imports SVG files into draw.io by converting SVG elements (paths, shapes, text, groups) into draw.io-compatible primitives. The conversion preserves visual properties (fill, stroke, opacity) and attempts to maintain structural hierarchy, allowing LLMs to reference SVG files and open them in draw.io for further editing and integration with other diagram elements.

Exposes diagram files (draw.io, Mermaid, CSV, SVG) as MCP resources, allowing LLMs to discover, list, and reference available diagrams in a project directory or workspace. The server scans the file system, indexes supported diagram formats, and provides resource URIs that LLMs can use to reference files in conversations and tool calls. Enables LLMs to maintain awareness of available diagrams without explicit file path specification.

Validates and parses draw.io XML files to extract diagram structure, elements, connections, and metadata. The parser reads draw.io's XML schema, validates file integrity, and provides structured access to diagram components (shapes, connectors, layers, styles). Enables LLMs to analyze existing diagrams, understand their structure, and make informed modifications or generate related diagrams.

Enables LLMs to generate draw.io diagrams programmatically by constructing draw.io XML from natural language descriptions or structured specifications. The LLM can describe diagram requirements (elements, connections, layout) and the MCP server translates these into valid draw.io XML with appropriate shapes, connectors, styling, and positioning. Allows LLMs to create diagrams directly without requiring users to manually draw them.

Exposes diagram operations (open, create, convert, validate) as MCP tools that LLMs can invoke via function calling. The server implements MCP tool schema with input/output specifications for each operation, allowing LLMs to call diagram functions with natural language intent translated to structured tool invocations. Enables seamless integration of diagram operations into LLM workflows and agent loops.

Handles reading and writing diagram files to disk in supported formats (draw.io XML, Mermaid, CSV, SVG). The server manages file I/O operations, ensures format-specific encoding and validation, and provides atomic write operations to prevent data loss. Enables LLMs to persist generated or modified diagrams to the file system for later retrieval and collaboration.

Exports diagrams from draw.io XML to multiple formats (Mermaid, SVG, PNG, PDF) for use in documentation, presentations, and external tools. The export process converts draw.io elements to target format specifications, applies format-specific styling and layout rules, and generates output files. Enables LLMs to generate diagrams in draw.io and export them to formats suitable for different use cases.