Skip to main content

Getting Started with DocuMCP

DocuMCP is a Model Context Protocol (MCP) server that gives your AI assistant the ability to analyze repositories, recommend static site generators, generate documentation structures, and deploy documentation sites. It runs as a background server and your AI client (Claude Desktop, Cursor, VS Code) communicates with it automatically — you interact entirely through natural language.

Prerequisites​

  • Node.js 20.0.0 or higher
  • npm (latest stable)
  • Git
  • One of: Claude Desktop, Cursor, or VS Code with GitHub Copilot

Step 0: Install DocuMCP and Connect Your AI Client​

Install DocuMCP​

Option A — npm global install (recommended)

npm install -g documcp

Option B — build from source (contributors / local development)

git clone https://github.com/tosin2013/documcp.git
cd documcp
npm install
npm run build

When using Option B, use the full path to dist/index.js in the MCP config below instead of "npx".

Connect to your AI client​

Choose the config for your client. Add it to the file shown, then restart the application.

Claude Desktop​

File location:

  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
  • Windows: %APPDATA%\Claude\claude_desktop_config.json
  • Linux: ~/.config/claude/claude_desktop_config.json
{
  "mcpServers": {
    "documcp": {
      "command": "npx",
      "args": ["documcp"]
    }
  }
}

Cursor​

File location: ~/.cursor/mcp.json (or add via Settings > MCP)

{
  "mcpServers": {
    "documcp": {
      "command": "npx",
      "args": ["documcp"]
    }
  }
}

VS Code (GitHub Copilot)​

Add to your settings.json:

{
  "mcp.servers": {
    "documcp": {
      "command": "npx",
      "args": ["documcp"]
    }
  }
}

Claude Code (CLI)​

If you are using Claude Code from the terminal, register DocuMCP with a single command.

npm install (global):

claude mcp add documcp -- npx documcp

git clone / local build (run from the cloned repo directory):

claude mcp add documcp -- node /path/to/documcp/dist/index.js

Example for a repo cloned to /root/documcp:

claude mcp add documcp -- node /root/documcp/dist/index.js

No restart needed — Claude Code picks up the new server immediately. Verify with:

claude mcp list

Verify the connection​

After restarting your AI client, ask it:

What tools does DocuMCP provide?

You should receive a list of available tools (analyze_repository, recommend_ssg, deploy_site, and others). If you do not, check that DocuMCP is installed and that the config file is saved correctly.

Tip: For the most detailed tool context, reference LLM_CONTEXT.md in your prompts once you have cloned or are working inside a DocuMCP-connected project:

@LLM_CONTEXT.md help me set up documentation for my TypeScript project

Step 1: Analyze Your Repository​

MCP tool: analyze_repository

Tell your AI assistant:

Analyze my repository at /path/to/your/project

Or with explicit parameters:

{
  "path": "/path/to/your/project",
  "depth": "standard"
}

DocuMCP returns:

  • Project structure — file counts, languages used, directory layout
  • Dependencies — production and development packages detected
  • Documentation status — existing docs, README, contributing guidelines
  • Smart recommendations — primary language, project type, team size estimates
  • Analysis ID — a unique identifier used in the next steps (e.g., analysis_abc123xyz)

Example response snippet:

{
  "id": "analysis_abc123xyz",
  "structure": {
    "totalFiles": 150,
    "languages": { ".ts": 45, ".js": 12, ".md": 8 },
    "hasTests": true,
    "hasCI": true
  },
  "recommendations": {
    "primaryLanguage": "typescript",
    "projectType": "library"
  }
}

Step 2: Get an SSG Recommendation​

MCP tool: recommend_ssg

Recommend a static site generator based on analysis_abc123xyz

Or with preferences:

{
  "analysisId": "analysis_abc123xyz",
  "preferences": {
    "ecosystem": "javascript",
    "priority": "features"
  }
}

DocuMCP uses its memory system — built from patterns across many past projects — to return confidence-scored recommendations:

{
  "recommended": "docusaurus",
  "confidence": 0.85,
  "reasoning": [
    "JavaScript/TypeScript ecosystem detected",
    "Modern React-based framework aligns with project stack"
  ],
  "alternatives": [
    {
      "name": "MkDocs",
      "score": 0.75,
      "pros": ["Simple setup", "Great themes"],
      "cons": ["Limited React component support"]
    }
  ]
}

Step 3: Generate Configuration​

MCP tool: generate_config

Generate a Docusaurus configuration for my project

Or explicitly:

{
  "ssg": "docusaurus",
  "projectName": "Your Project",
  "projectDescription": "Your project description",
  "outputPath": "/path/to/your/repository"
}

This creates a ready-to-use docusaurus.config.js (or equivalent) tailored to your project.

Step 4: Set Up Documentation Structure​

MCP tool: setup_structure

Set up a Diataxis documentation structure for my project using Docusaurus

Or explicitly:

{
  "path": "/path/to/your/repository/docs",
  "ssg": "docusaurus",
  "includeExamples": true
}

This creates four Diataxis-compliant sections:

  • Tutorials — learning-oriented guides for skill acquisition
  • How-to Guides — problem-solving guides for specific tasks
  • Reference — information-oriented content for lookup
  • Explanation — understanding-oriented conceptual content

Step 5: Deploy​

MCP tool: deploy_site

To deploy to GitHub Pages:

Deploy my documentation to GitHub Pages
{
  "projectPath": "/path/to/your/repository",
  "target": "github-pages",
  "ssg": "docusaurus"
}

To deploy to Vercel:

Deploy my documentation to Vercel
{
  "projectPath": "/path/to/your/repository",
  "target": "vercel",
  "ssg": "docusaurus"
}

deploy_site generates:

  • A GitHub Actions workflow file (.github/workflows/deploy-github-pages.yml or deploy-vercel.yml)
  • OIDC authentication with minimal permissions
  • A setup checklist (VERCEL_SETUP.md for Vercel) committed to your repo

Back-compat note: The older deploy_pages tool still works and maps to deploy_site with target=github-pages. It will be removed in v1.1.0.

What You Learned​

  • What DocuMCP is and how it integrates with your AI client
  • How to install and connect DocuMCP (npm or git clone)
  • How to verify the MCP server is running
  • How to run the core workflow: analyze → recommend → configure → structure → deploy

Next Steps​