Skip to main content
Glama

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.

PyPI Python 3.10+ License: MIT


Tools

Tool

Description

parse_pdf

Extract text + bounding boxes (x, y, width, height in PDF points) from a PDF

batch_parse_pdfs

Parse every PDF in a folder; write JSON + screenshots per file

screenshot_pdf

Render pages as base64 PNG images

cited_screenshot

Render a page with highlight boxes drawn over every text item

search_pdf

Find a phrase and return all matching positions with coordinates

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

Available Tools

5 tools
batch_parse_pdfsA

Parse all PDFs in a folder. For each PDF, writes to <output_folder>//: pages.json – text + bounding boxes per page summary.txt – plain text of the whole document page_N.png – raw page screenshot page_N_cited.png – screenshot with bounding-box highlights

Returns a JSON summary of successes and any errors.

Inputs: folder_path (required), max_workers, ocr_enabled, output_folder. Returns: result

cited_screenshotA

Render a single PDF page as a PNG with bounding-box highlights drawn over every extracted text item (or only items matching search_phrase).

Returns JSON: { "page_num", "width_px", "height_px", "items_highlighted", "image_base64" }

Inputs: pdf_path (required), dpi, page_number, search_phrase, +1 more. Returns: result

parse_pdfA

Parse a PDF and return structured text plus bounding boxes for every text item on every page.

Bounding-box coordinates are in PDF points (1 pt = 1/72 inch), with the origin at the top-left corner of each page.

Returns JSON: { "pages": [ { "page_num", "width_pts", "height_pts", "text", "text_items": [ { "text", "x", "y", "width", "height", "font_name", "font_size", "confidence" } ] } ] }

Inputs: pdf_path (required), dpi, pages, ocr_enabled. Returns: result

screenshot_pdfA

Render PDF pages as PNG images (base64-encoded).

Returns JSON list of: { "page_num", "width_px", "height_px", "image_base64" }

Inputs: pdf_path (required), dpi, page_numbers. Returns: result

search_pdfA

Search for a phrase in a PDF and return every matching text item with its page number and bounding-box coordinates.

Returns JSON: { "phrase", "match_count", "matches": [ { "page_num", "text", "x", "y", "width", "height", "font_size" } ] }

Inputs: phrase (required), pdf_path (required), case_sensitive. Returns: result

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A4.1/5.0

    Scored across 5 tools

    Disambiguation5/5

    Each tool has a clearly distinct purpose: rendering pages, extracting text with coordinates, batch processing, highlighted screenshots, and text search. No two tools serve the same function, even though two produce images, the context and output differ significantly.

    Naming Consistency4/5

    All tool names are in snake_case and mostly follow a verb_noun pattern (screenshot_pdf, parse_pdf, batch_parse_pdfs, search_pdf). 'cited_screenshot' is the exception as it is a descriptive noun phrase rather than a verb_noun, but it still fits the overall naming style.

    Tool Count5/5

    Five tools cover the core PDF parsing and rendering workflow without being excessive. Each tool addresses a distinct need, from single-page screenshots to batch processing, making the tool count well-scoped for the server's purpose.

    Completeness5/5

    The tool set covers the primary PDF operations: text extraction (parse_pdf), visual rendering (screenshot_pdf), highlighting (cited_screenshot), searching (search_pdf), and batch processing (batch_parse_pdfs). This is a complete lifecycle for the server's stated PDF parsing and analysis purpose.

    Maintenance

    ActivityInactive
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      A
      quality
      C
      maintenance
      An MCP server for reading, rendering, and searching PDF files, specifically optimized for LLMs to extract text, tables, and technical diagrams. It enables metadata retrieval, multi-format text extraction, and page-to-image rendering using PyMuPDF.
      5
      83
      MIT
    • F
      license
      Not graded
      quality
      D
      maintenance
      MCP server for extracting text from PDF files, supporting local files and URLs.
      -
    • A
      license
      A
      quality
      D
      maintenance
      This MCP server enables AI agents to view PDFs as accessible HTML with bounding-box citations, and provides tools for layout-aware parsing, schema extraction, cross-document Q&A, and PDF rendering.
      27
      MIT
    • F
      license
      Not graded
      quality
      D
      maintenance
      A local MCP server that extracts text-layer content from PDF files, enabling AI agents to inspect, extract text, outlines, and page content.
      -