Document MCP Server

A Model Context Protocol (MCP) server designed to efficiently manage component documentation with AI-powered updates and token optimization.

Features

• 🚀 Component Scanning: Automatically scan project components folder and generate structured information

• 🔐 MD5 Hashing: Generate MD5 hashes for files to track changes

• 📋 Component Queries: Quickly retrieve specific component details

• ✏️ Precision Updates: Update only description or props sections, saving significant tokens

• 🤖 AI-Friendly: Structured updates prevent AI hallucinations and unnecessary modifications

• 📊 Token Optimization: Save ~75% tokens compared to traditional full-file regeneration

Related MCP server: project-graph-mcp

Installation

NPM Package

# Install globally
npm install -g @tianyi-li/document-mcp

# Install as project dependency
npm install @tianyi-li/document-mcp

Development Setup

# Clone and setup
git clone https://github.com/TianyiLi/document-mcp.git
cd document-mcp
pnpm install
pnpm run build

Quick Start

1. Basic Usage

# Run the MCP server
document-mcp

# Or use npx/pnpx
npx @tianyi-li/document-mcp

2. IDE Integration

Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "document-mcp": {
      "command": "npx",
      "args": ["@tianyi-li/document-mcp"]
    }
  }
}

Claude Code

Option 1: Through Claude Desktop (Automatic)

Claude Code automatically detects MCP servers configured in Claude Desktop - no additional setup required.

Option 2: Direct CLI Configuration

Configure directly via Claude Code CLI:

# Using npx (recommended)
claude mcp add document-mcp -- npx @tianyi-li/document-mcp

# Using global installation
claude mcp add document-mcp -- document-mcp

# List configured servers
claude mcp list

# Remove if needed
claude mcp remove document-mcp

Project Documentation (CLAUDE.md)

Add to your project's CLAUDE.md for team reference:

# Document MCP Server

This project uses Document MCP Server for efficient component documentation management.

## Setup
```bash
# Quick setup
npm install -g @tianyi-li/document-mcp
claude mcp add document-mcp -- document-mcp

Available Tools

• init_components - Scan all project components

• get_component_info - Get specific component details

• update_component_description - Update description only (saves ~75% tokens)

• update_component_props - Update props only (saves ~75% tokens)

• update_component_content - Batch update multiple sections

• compare_md5 - Verify file changes

Workflow

1. init_components → Understand project structure

2. update_component_* → Make targeted updates

3. compare_md5 → Verify changes

## API Reference

### Reading Tools

#### `init_components`
Scan components folder and generate structured JSON data.

**Parameters:**
- `projectRoot` (optional): Project root path, defaults to current directory

**Returns:**
```json
[
  {
    "name": "Button",
    "path": "components/Button", 
    "description": "A reusable button component",
    "props": "Props information",
    "md5": "a1b2c3d4e5f6..."
  }
]

get_component_info

Get detailed information for a specific component.

Parameters:

• componentName (required): Component name

• projectRoot (optional): Project root path

Returns: Single component JSON object

Update Tools ⭐ Core Value

update_component_description

Precisely update component description section - best choice for token saving.

Parameters:

• componentName (required): Component name

• description (required): New component description

• projectRoot (optional): Project root path

Usage: AI agents can update only the description section without regenerating entire files

update_component_props

Precisely update component Props section with Markdown table support.

Parameters:

• componentName (required): Component name

• props (required): New Props content (supports Markdown format)

• projectRoot (optional): Project root path

Example Props Format:

| Property | Type | Required | Default | Description |
|----------|------|----------|---------|-------------|
| variant | "primary" \| "secondary" | No | "primary" | Button style |
| size | "small" \| "medium" \| "large" | No | "medium" | Button size |
| disabled | boolean | No | false | Whether disabled |

update_component_content

Batch update component description and/or Props with partial update support.

Parameters:

• componentName (required): Component name

• description (optional): New component description

• props (optional): New Props content

• projectRoot (optional): Project root path

Usage: Can update multiple sections simultaneously or individually

Validation Tools

calculate_md5

Calculate MD5 hash for specified file.

Parameters:

• filePath (required): File path for MD5 calculation

Returns: MD5 hash string

compare_md5

Compare current file MD5 with expected value.

Parameters:

• filePath (required): File path to check

• expectedMd5 (required): Expected MD5 hash value

Returns:

{
  "currentMd5": "current file MD5",
  "expectedMd5": "expected MD5", 
  "isMatch": true/false,
  "hasChanged": true/false
}

Use Cases

Reading Phase

1. Automated Documentation Management: AI agents quickly understand project component structure

2. Component Information Queries: Avoid scanning entire project structure, saving tokens

3. File Change Tracking: Ensure document content consistency through MD5 comparison

Update Phase ⭐ Core Value

4. Precision Document Updates: Update only necessary parts (description or Props), dramatically saving tokens

5. Prevent AI Hallucinations: Structured updates avoid AI adding/removing unnecessary items

6. Incremental Maintenance: Modify based on existing content rather than complete rewrites

7. Consistency Guarantee: Use standard templates and formats for unified document style

Workflow Example

1. init_components          → Understand all components
2. get_component_info       → Get specific component current state
3. update_component_description → Update only description section
4. compare_md5              → Verify update success

Token Savings

• Traditional Method: AI reads full file + regenerates complete file = ~2000 tokens

• Using This Tool: Get existing info + precise partial updates = ~500 tokens

• Savings Rate: ~75% token reduction

Project Structure

This tool assumes your project has the following structure:

project-root/
├── components/
│   ├── Button/
│   │   └── README.md
│   ├── Input/
│   │   └── README.md
│   └── ...
└── ...

Each component folder should contain a README.md file with component description and props information.

Configuration Examples

Using pnpx/npx (Recommended)

{
  "mcpServers": {
    "document-mcp": {
      "command": "pnpx",
      "args": ["@tianyi-li/document-mcp"],
      "cwd": "/path/to/your/project"
    }
  }
}

Direct Binary Execution

{
  "mcpServers": {
    "document-mcp": {
      "command": "/path/to/node_modules/.bin/document-mcp"
    }
  }
}

Global Installation

# Install globally
npm install -g @tianyi-li/document-mcp

# Use in Claude Desktop config
{
  "mcpServers": {
    "document-mcp": {
      "command": "document-mcp"
    }
  }
}

Development

Build from Source

git clone https://github.com/TianyiLi/document-mcp.git
cd document-mcp
pnpm install
pnpm run build

Available Scripts

pnpm run build    # Compile TypeScript
pnpm run dev      # Watch mode compilation
pnpm start        # Run the server

Best Practices

1. First Use: Run init_components to scan project components

2. Change Tracking: Use compare_md5 to check for file modifications

3. Specific Queries: Use get_component_info for individual component details

4. Regular Sync: Periodically run init_components to keep data current

5. Token Optimization: Use precise update tools instead of full rewrites

Requirements

• Node.js >= 18.0.0

• Components organized in /components folder structure

• Each component should have a README.md file

License

ISC License

Contributing

Issues and pull requests are welcome on GitHub.

Repository

• Homepage: https://github.com/TianyiLi/document-mcp

• Issues: https://github.com/TianyiLi/document-mcp/issues

• NPM: @tianyi-li/document-mcp