mcp_pdf_processor

The PDF Processor MCP server enables Claude to fetch PDFs from URLs, extract text, and identify LaTeX equations for downstream analysis. It exposes MCP-driven commands under the PDF_TOOLS namespace that let you fetch a PDF, process it (with optional LaTeX extraction), and then read the processed content. Typical use cases include retrieving a document, extracting mathematical expressions in LaTeX form, and summarizing or analyzing the content for further insights. When registered with Claude, you can invoke these tools directly in conversations to perform end-to-end PDF processing without leaving the chat.

Prerequisites:

• Python 3.9 or higher
• pip (Python package manager)
• Optional: MCP CLI tools if you plan to integrate with Claude Desktop/Claude Code

1. Install the package in editable mode (from the repository root):

pip install -e .

2. (Optional for Claude Desktop/Code) Install MCP CLI tools:

pip install "mcp[cli]"

3. Install or run the server locally:

# Run the server directly (standalone mode)
python pdf_tool_server.py

4. If you want to register/install with Claude Desktop/Claude Code via MCP CLI:

# Install the server using the MCP CLI tool
mcp install /path/to/pdf_tool_server.py --with-editable /path/to/mcp_pdf_processor

Example with your repo cloned at ~/mcp_pdf_processor:

mcp install ~/mcp_pdf_processor/pdf_tool_server.py --with-editable ~/mcp_pdf_processor

5. For development with the MCP Inspector:

mcp dev /path/to/pdf_tool_server.py --with-editable /path/to/mcp_pdf_processor

Note: Ensure that the PD F processing dependencies (pymupdf, torch, pix2tex, etc.) specified in pyproject.toml are installed as needed for full functionality.

Environment variables:

• OUTPUT_DIR controls where processed PDFs are stored. If not set, the default is llm_output.
• PYTHONPATH should point to the directory containing the mcp_pdf_processor package to ensure imports resolve correctly.

Common issues:

• Ensure Python 3.9+ is installed; some dependencies may require newer versions.
• If LaTeX extraction relies on optional components (torch, pix2tex), install them or disable those features if not needed.
• When using Claude, the server name (PDF_TOOLS in instructions) should map to the registered mcpServer key (pdf_tool_server in this config).

Configuration tips:

• You can adjust OUTPUT_DIR to a writable path in your environment to avoid permission errors.
• If you encounter import errors, verify PYTHONPATH includes the root of the mcp_pdf_processor package.