Model Context Protocol • Local Stdio Server

PDFLince MCP Server

Connect Claude Desktop, Cursor, Antigravity, Windsurf, or VS Code to a local, zero-cloud PDF engine. Inspect, merge, split, extract, reorder, rotate, crop, compress, and convert PDFs directly on your filesystem.

10MCP Tools0External Requests100%Local Execution< 30msTypical Overhead

Client Configuration

Add the server definition to your MCP client configuration file.

claude_desktop_config.json

{
  "mcpServers": {
    "pdflince": {
      "command": "npx",
      "args": [
        "-y",
        "pdflince-mcp@latest"
      ]
    }
  }
}

Direct CLI / stdio invocation:

Tool Reference

Ten structured MCP tools mirroring every operation and option available in the PDFLince interface.

Inspect PDF

pdflince_inspect

Read

Reads a local PDF without modifying it. Returns total page count, per-page dimensions, rotation angles, orientation, and document metadata.

Example Prompt

“Inspect financial-report.pdf and return the page count, dimensions, and author metadata.”

Parameters

Name	Type	Required	Description
filePath	string	Yes	Absolute or relative path to the local PDF file.

Strict Data Sovereignty

Process contracts, financial statements, medical records, and internal documentation locally without exposing files to third-party APIs.

Full UI Feature Parity

Exposes all 9 operations from the PDFLince web application—merge, compress, split, extract, reorder, rotate, crop, PDF-to-images, and images-to-PDF—plus deep document inspection.

Universal Host & Model Support

Works with Claude Desktop, Cursor, Antigravity, Windsurf, Cline, Roo Code, Zed, and Open WebUI—across Claude, Grok, GPT-4o, or local Ollama models (Nous Hermes, DeepSeek).

Zero network egress or telemetry: all operations execute inside a local Node.js or Bun process.

Designed for regulated workflows and strict data governance (GDPR, HIPAA, SOC 2).

Non-destructive file handling: transformation tools write to explicit output paths without mutating source files.

Frequently Asked Questions

How does the MCP server operate offline?

The `pdflince-mcp` binary runs as a local child process spawned by your MCP host (Claude Desktop, Cursor, Antigravity) and communicates over stdio JSON-RPC. It opens no network ports and makes no external HTTP requests.

What are the runtime requirements?

Node.js 18+ (Node 20+ recommended) or Bun. No Python environment or external cloud API keys are required.

Does it work with local LLMs (Ollama, Nous Hermes, DeepSeek)?

Yes. Because PDFLince implements the standard stdio Model Context Protocol, it is model-agnostic. You can pair it with local open-weights models (Nous Hermes 3, DeepSeek-R1/V3, Llama 3 via Ollama, LM Studio, or vLLM) as well as commercial models (Claude, Grok, GPT-4o).

Are source PDF files modified in place?

No. Every write operation requires an explicit `outputPath` or `outputDirectory`, leaving original source files untouched.

Zero-cloud PDF processing for your IDE and AI agents in under a minute.

Open Web Application View Source on GitHub