PDF Reader MCP Server (@sylphlab/pdf-reader-mcp)

@sylphlab

代码仓库主页

基本信息

分类

文件与存储

许可证

MIT

传输方式

stdio

作者

sylphlab

关于 PDF Reader MCP Server

📄 The PDF intelligence layer for AI agents — Agent Document Twin, evidence-first extraction, visual crops, OCR provenance, trust reports, and benchmark-gated releases. MCP server for Claude, Cursor, VS Code, and any MCP client.

#nodejs #pdf #typescript #mcp #stdio #pdf-reader #pdf-parser#pdf-parse#ai-agent#model-content-protocol#llm-tool

PDF Reader MCP Server 配置

使用下面的配置,将此服务器添加到你的 MCP 客户端。

{
  "mcpServers": {
    "pdf-reader-mcp-sylphlab": {
      "command": "npx",
      "args": [
        "@sylphx/pdf-reader-mcp"
      ]
    }
  }
}

PDF Reader MCP Server 工具

工具列表暂未提供

当前收录暂时没有可展示的工具列表，不代表服务没有工具。请查阅项目文档了解支持的能力。

点击上方按钮,在临时沙箱中运行此服务器并实时获取工具列表。

概览

What is PDF Reader MCP Server (@sylphlab/pdf-reader-mcp)?

PDF Reader MCP Server is an MCP server that provides an Agent Document Twin for PDFs – a linked, source-backed representation that AI agents can inspect, search, verify, crop, OCR, enrich, cite, and read with confidence. It is designed for agents that need evidence, not just extracted text, and handles digital, scanned, and mixed PDFs with tables, hidden text, annotations, and complex layouts.

How to use PDF Reader MCP Server (@sylphlab/pdf-reader-mcp)?

Install the npm package @sylphx/pdf-reader-mcp (Node.js ≥22.13 required) via npx, or use the Docker image from GitHub Container Registry. Add it to Claude Desktop config, Claude Code, or any MCP client. The default package works without OCR models, vision models, or cloud credentials. Use the read_pdf tool first with only sources to auto-profile and read the PDF, or use explicit include_* options for precise control. Also supports search_pdf and pdf_evidence tools.

Key features of PDF Reader MCP Server (@sylphlab/pdf-reader-mcp)

Agent Document Twin with lossless, visual, semantic, evidence, and agent layers.
One smart read_pdf tool that auto-profiles and extracts with evidence.
OCR adapter for scanned pages via configured providers (e.g. Tesseract).
Visual crops, region analysis, and provider‑normalized evidence.
Trust reports for hidden text, unsafe links, spoofing, and prompt injection.
Accessibility reports for tagged‑PDF coverage, headings, links, and forms.
Deterministic quality benchmarks and release gates for shipped proof.

Use cases of PDF Reader MCP Server (@sylphlab/pdf-reader-mcp)

Read a PDF with markdown, tables, chunks, and page numbers for summarization or RAG.
Search within a PDF and then visually verify the source region before citing.
Extract tables from scanned or selectable-text PDFs with cell geometry and confidence.
Protect agents from misleading PDF content using trust and accessibility reports.
Ship a PDF intelligence pipeline with reproducible quality benchmarks and release artifacts.

常见问题

What is the Agent Document Twin?

It is a multi‑layer output (lossless, visual, semantic, evidence, agent) that keeps the PDF readable by agents while preserving the provenance needed to verify answers – page numbers, bounding boxes, crop IDs, and extraction methods.

Does it work without any external AI models?

Yes. The default package works out of the box for selectable‑text PDFs, rendering, crops, trust reports, and accessibility reports. OCR for scanned pages and visual enrichment are optional provider‑enabled features configured by the deployment environment.

What are the runtime requirements?

Node.js ≥22.13 is required. No OCR models, vision models, Ollama, LM Studio, or cloud credentials are needed by default.

How is OCR configured?

Set environment variables such as `MCP_PDF_OCR_COMMAND` and `MCP_PDF_OCR_ARGS_JSON` to point to a local OCR command (e.g. Tesseract). Provider URLs and executables are controlled by the environment, not by request payloads.