Outline Wiki MCP Server

Instructions

Implementation Reference

• src/lib/handlers/smart.ts:193-204 (handler)

  MCP tool handler for generate_diagram. Checks if brain is enabled, calls brain.generateDiagram, and returns the Mermaid diagram code with rendering instructions.

  async generate_diagram(args: { description: string }) {
    if (!brain.isEnabled()) {
      return { error: ERROR_MESSAGES.SMART_FEATURES_DISABLED };
    }
  
    const diagram = await brain.generateDiagram(args.description);
  
    return {
      diagram,
      note: 'Copy this Mermaid code into an Outline document to render the diagram.',
    };
  },

• src/lib/schemas.ts:164-166 (schema)

  Zod schema defining the input for generate_diagram tool: requires a description string.

  export const generateDiagramSchema = z.object({
    description: z.string().min(1, 'Description is required'),
  });

• src/lib/tools.ts:225-229 (registration)

  Tool registration in the allTools array, converting the Zod schema to JSON schema for MCP.

  createTool(
    'generate_diagram',
    'Generate a Mermaid diagram from a text description.',
    'generate_diagram'
  ),

• src/lib/brain/index.ts:178-181 (helper)

  Brain class method that checks if enabled and delegates diagram generation to LlmProcessor.

  async generateDiagram(description: string): Promise<string> {
    this.checkEnabled();
    return this.processor.generateMermaid(description);
  }

• src/lib/brain/processor.ts:81-87 (helper)

  Core implementation in LlmProcessor using OpenAI chat completion with a specific prompt to generate Mermaid code from description.

    async generateMermaid(description: string): Promise<string> {
      const prompt = `Convert the following process or flow description into a Mermaid.js diagram.
  Output ONLY the Mermaid code block, no explanation.
  Use appropriate diagram type (flowchart, sequence, etc.) based on the content.`;
  
      return this.complete(prompt, description);
    }