liteparse-mcp

Fast, local PDF parsing as an MCP server — text extraction, bounding boxes, OCR, and visual citations. No cloud. No API key. Powered by LiteParse.

Tools

Bounding-box coordinates are in PDF points (1 pt = 1/72 in), origin top-left. To convert to pixels: px = pt × (dpi / 72).

Related MCP server: pdf-reader-mcp

Install

pip install liteparse-mcp

Usage

Claude Desktop

Add to ~/AppData/Roaming/Claude/claude_desktop_config.json (Windows) or ~/Library/Application Support/Claude/claude_desktop_config.json (macOS):

{
  "mcpServers": {
    "liteparse": {
      "command": "liteparse-mcp"
    }
  }
}

Or with the explicit Python path (if liteparse-mcp is not on PATH):

{
  "mcpServers": {
    "liteparse": {
      "command": "python",
      "args": ["-m", "liteparse_mcp"]
    }
  }
}

Restart Claude Desktop — the five tools appear automatically.

Claude Code

claude mcp add liteparse -- python -m liteparse_mcp

HTTP / SSE (for remote agents or testing)

liteparse-mcp --http
# Server listens on http://127.0.0.1:8765

Example agent prompts

• "Parse report.pdf and show me where 'efficacy' appears with bounding boxes"

• "Get a cited screenshot of page 3 of study.pdf"

• "Batch parse every PDF in my Downloads folder and save the output"

• "Search safety_data.pdf for 'adverse event' and list the page numbers"

Outputs (batch mode)

For each PDF, batch_parse_pdfs writes:

<output_folder>/
  <stem>/
    pages.json          # structured JSON: page text + TextItem bounding boxes
    summary.txt         # plain text of the whole document
    page_1.png          # raw page screenshot
    page_1_cited.png    # screenshot with bounding-box highlights
    ...
  batch_report.json     # overall success / error summary

Requirements

• Python ≥ 3.10

• liteparse ≥ 2.0.0 (Rust-based; wheels available for Windows, macOS, Linux)

• fastmcp ≥ 2.0.0

No Tesseract installation required for text-based PDFs. For scanned PDFs with ocr_enabled=true, Tesseract is used automatically if available on PATH.

License

MIT