Design Philosophy

The PDF Reader MCP Server is built upon several core principles:

1. Security First:

   • Context Confinement: The absolute primary goal. All local file access must be restricted to the directory (and its subdirectories) where the server process is launched. This prevents the AI agent from accessing unintended files on the user's system.
   • Path Validation: Rigorous validation of all incoming paths using a dedicated resolvePath function ensures they are relative and resolve within the designated project root.
   • No Arbitrary Execution: The server only performs PDF reading operations, not arbitrary file system modifications or command execution.

2. Efficiency & Resourcefulness:

   • Structured Data: Instead of sending potentially huge raw PDF content (which is often impractical for LLMs), the server extracts specific, structured information (text, metadata, page count).
   • Targeted Extraction: Allows requesting text from specific pages, minimizing the amount of data transferred and processed.
   • Asynchronous Operations: Uses Node.js async I/O to avoid blocking the event loop during file access and PDF parsing.

3. Simplicity & Ease of Integration:

   • Single Tool Focus: Consolidates functionality into a single read_pdf tool with clear parameters, making it easier for AI agents to learn and use.
   • Standard MCP: Leverages the @modelcontextprotocol/sdk for standard communication and error handling.
   • Clear Schemas: Uses Zod for defining and validating input, providing clear contracts for tool usage.
   • Multiple Invocation Methods: Supports easy use via npx or Docker for straightforward deployment in various MCP host environments.

4. Minimalism & Reliability:

   • Minimal Dependencies: Relies primarily on the robust and widely-used pdfjs-dist library for core PDF parsing, minimizing external failure points.
   • Clear Error Reporting: Provides specific error messages when processing fails for a source, allowing the agent to understand the issue.