# Nutrient Document Engine MCP Server

![Document workflows using natural language](https://raw.githubusercontent.com/PSPDFKit/nutrient-document-engine-mcp-server/main/resources/readme-header.png)

[![npm](https://img.shields.io/npm/v/%40nutrient-sdk/document-engine-mcp-server)](https://www.npmjs.com/package/@nutrient-sdk/document-engine-mcp-server)

**Connect AI agents to document processing.**

This MCP server exposes document processing capabilities through natural language, supporting both direct human interaction and AI agent tool calling.

## Which MCP Server Should I Use?

| Server | Best for | Deployment | Core capabilities |
|---|---|---|---|
| [Nutrient DWS MCP Server](https://github.com/PSPDFKit/nutrient-dws-mcp-server) | Cloud document workflows | Nutrient-hosted API (API key) | Convert, OCR, redact, sign, extract, watermark, automation |
| [Nutrient Document Engine MCP Server](https://github.com/PSPDFKit/nutrient-document-engine-mcp-server) | Self-hosted document workflows | On-prem/private cloud | Document processing with deployment control and data residency |

You are in the **Document Engine MCP Server** repo. Choose this when you need self-hosted/on-prem document workflows and deployment control.

- Ecosystem overview: [Nutrient AI Infrastructure](https://www.nutrient.io/ai/infrastructure/)
- Product landing page: [Nutrient MCP Server](https://www.nutrient.io/mcp-server-pdf-automation-llm/)

## Example Conversations

Once configured, you (or your agent) can interact with documents through natural language:

**You:** *"What documents do I have available?"*  
**AI:** *"I can see you have 3 documents: annual-report.pdf, contract-draft.pdf, and tax-form-2023.pdf. Would you like me to analyze any of these?"*

**You:** *"Extract all the names and phone numbers from these invoices"*  
**AI:** *"I found 12 contacts: John Smith (555-0123), Sarah Johnson (555-0456)..."*

**You:** *"Redact all social security numbers from these tax forms"*  
**AI:** *"I've identified and redacted 3 SSNs across 2 documents. The redacted versions are ready."*

**You:** *"Split this 50-page contract into separate sections"*  
**AI:** *"I've split the contract into 5 sections: Terms (pages 1-8), Payment (pages 9-15)..."*

## Setup with Claude Desktop

![Document Engine MCP Server + Claude Desktop](https://raw.githubusercontent.com/PSPDFKit/nutrient-document-engine-mcp-server/main/resources/claude-document-engine-mcp.gif?raw=true)

### Prerequisites
- **Claude Desktop** - [Download from Anthropic](https://claude.ai/download).
- **Docker Compose** - [Download from Docker](https://docs.docker.com/compose/). For running Nutrient Document Engine locally.
- **Node.js 18+** - This MCP server runs via `npx`.

### Licensing

Nutrient Document Engine runs with an evaluation license by default. Please see [Removing Evaluation Limitations](#removing-evaluation-watermarks-and-limitations) section on how to remove the evaluation limitations.

### 1. Start Nutrient Document Engine
Open a terminal window, then run:

```bash
git clone https://github.com/PSPDFKit/nutrient-document-engine-mcp-server.git
cd nutrient-document-engine-mcp-server
docker-compose up -d
```

### 2. Connect Claude Desktop

Add this to your Claude Desktop config (Settings → Developer → Edit Config):

```json
{
  "mcpServers": {
    "nutrient-document-engine": {
      "command": "npx",
      "args": ["-y", "@nutrient-sdk/document-engine-mcp-server"],
      "env": {
        "DASHBOARD_USERNAME": "admin",
        "DASHBOARD_PASSWORD": "password",
        "DOCUMENT_ENGINE_BASE_URL": "http://localhost:5000",
        "DOCUMENT_ENGINE_API_AUTH_TOKEN": "secret"
      }
    }
  }
}
```

### 3. Restart Claude Desktop

### 4. Upload and Process Documents
1. Open your web browser and go to http://localhost:5100/dashboard.
2. Upload the documents you want to work with by using the drag-and-drop interface.
3. Switch back to Claude Desktop and ask questions like:
   *"List my documents and extract text from the contract"*

## HTTP Transport Security

The optional HTTP transport binds to `127.0.0.1` by default. A non-loopback `MCP_HOST` requires
`MCP_HTTP_AUTH_TOKEN`; the server refuses to start without a nonempty token. Clients must send the
token on every `/mcp` request as `Authorization: Bearer <token>`. If a token is set for a loopback
HTTP server, `/mcp` requires it there too. This token does not protect `/health` or `/dashboard`;
dashboard access uses `DASHBOARD_USERNAME` and `DASHBOARD_PASSWORD`.

```bash
# Local-only HTTP transport; no MCP bearer token required.
MCP_TRANSPORT=http npx @nutrient-sdk/document-engine-mcp-server

# Network-accessible HTTP transport; bearer authentication is required.
MCP_TRANSPORT=http MCP_HOST=0.0.0.0 MCP_HTTP_AUTH_TOKEN=replace-with-a-secret \
  npx @nutrient-sdk/document-engine-mcp-server
```

The default stdio transport is unchanged and never requires `MCP_HTTP_AUTH_TOKEN`.

## Features

This MCP server provides document processing tools in these areas:

**Content Access**: Extract text, search documents, render pages as images, extract form data and table structures.

**Document Editing**: Split documents, add watermarks, duplicate documents, fill PDF forms.

**Security**: Create and apply redactions for sensitive data using preset patterns (SSNs, credit cards, etc.) or custom regex.

**Annotations**: Add, read, and delete document annotations including highlights, notes, and stamps.

**System**: Document discovery, metadata retrieval, and health monitoring.

## Use with Other AI Platforms

This MCP server works with any AI platform that supports the Model Context Protocol:

- **Claude Desktop** - Quick setup with the configuration above.
- **LangGraph** - Integrate with [LangGraph agents](examples/langgraphjs) for complex workflows.
- **OpenAI APIs** - Connect through MCP-compatible frameworks.
- **Custom agents** - Any MCP-compatible AI system.

## Examples and Documentation

### Examples
- [LangGraph Document Processing](examples/langgraphjs) - Workflow examples for contract processing, form filling, and report generation.
- [OpenAI Agents SDK Document Processing](examples/openai-agents) - Workflow examples for contract processing, form filling, and report generation.
- [Document Chat Interface](examples/document-chat) - Web application with React and LangGraph.
- [Procurement AI Agent](examples/procurement-agent-ts) - An AI Agent written with Langgraph.js, that classifies and collates POs, invoices, and payment documents into groups and flags for human review

### Documentation
- [Configuration Guide](docs/configuration.md) - Environment variables, transport modes, dashboard.
- [Features Reference](docs/features.md) - Detailed description of all document processing tools.
- [Evaluation Framework](evaluation) - Testing and benchmarking tools.

## Removing Evaluation Watermarks and Limitations

Document Engine runs in evaluation mode with limitations by default. To remove them:

1. **Contact Sales to purchase a license** from [Nutrient](https://www.nutrient.io/contact-sales)
2. **Configure your license** by setting the `ACTIVATION_KEY` environment variable in your [Docker Compose configuration](docker-compose.yml) or Document Engine setup
3. **Restart Document Engine** to apply the license

For detailed licensing configuration, see the [Document Engine licensing guide](https://www.nutrient.io/guides/document-engine/about/licensing/).

## Server Based Document Engine Setups

**On-Premise Installation:** Follow the [Nutrient Document Engine installation guide](https://www.nutrient.io/getting-started/document-engine/).

**Managed Cloud:** Use a hosted Document Engine instance - [setup guide](https://www.nutrient.io/guides/document-engine/deployment/managed-document-engine/).

## Contributing

We welcome contributions! See [CONTRIBUTING.md](CONTRIBUTING.md) for development setup, testing guidelines, and how to submit pull requests.

## License

MIT License - see [LICENSE](LICENSE) file for details.