PDF Redaction MCP Server

Instructions

Implementation Reference

• src/redact_mcp/server.py:23-80 (handler)

  The core handler function for the 'load_pdf' tool. Decorated with @mcp.tool for automatic registration. Loads the PDF using PyMuPDF (fitz), validates file existence, extracts and returns full text content from all pages while storing the document object in a global dictionary for subsequent operations.

  @mcp.tool
  async def load_pdf(
      pdf_path: Annotated[str, Field(description="Path to the PDF file to load")],
      ctx: Context
  ) -> str:
      """Load a PDF file and make it available for redaction.
      
      This tool loads a PDF file into memory and extracts its text content
      for review. The PDF remains loaded for subsequent redaction operations.
      
      Args:
          pdf_path: Path to the PDF file to load
          ctx: MCP context for logging
          
      Returns:
          The full text content of the PDF
          
      Raises:
          ToolError: If the file doesn't exist or cannot be opened
      """
      try:
          path = Path(pdf_path).resolve()
          
          await ctx.info(f"Loading PDF from: {path}")
          
          if not path.exists():
              raise ToolError(f"PDF file not found: {path}")
          
          if not path.is_file():
              raise ToolError(f"Path is not a file: {path}")
              
          # Open the PDF
          doc = fitz.open(str(path))
          
          # Store the document for later use
          _loaded_pdfs[str(path)] = doc
          
          # Initialize redaction tracking for this PDF
          if str(path) not in _applied_redactions:
              _applied_redactions[str(path)] = []
          
          # Extract text from all pages
          text_content = []
          for page_num, page in enumerate(doc, start=1):
              page_text = page.get_text()
              text_content.append(f"--- Page {page_num} ---\n{page_text}")
          
          full_text = "\n\n".join(text_content)
          
          await ctx.info(f"Successfully loaded PDF with {len(doc)} pages")
          
          return full_text
          
      except ToolError:
          raise
      except Exception as e:
          await ctx.error(f"Failed to load PDF: {str(e)}")
          raise ToolError(f"Failed to load PDF: {str(e)}")

• src/redact_mcp/server.py:25-26 (schema)

  Pydantic schema definition for the tool input using Annotated and Field, specifying the pdf_path parameter with description.

  pdf_path: Annotated[str, Field(description="Path to the PDF file to load")],
  ctx: Context

• src/redact_mcp/server.py:23-23 (registration)

  The @mcp.tool decorator registers the load_pdf function as an MCP tool with FastMCP instance.

  @mcp.tool