Extract text, metadata, and tables from PDFs (local or URL) via @sylphx/pdf-reader-mcp (npx)

Recipe: Extract text, metadata, and tables from PDFs via @sylphx/pdf-reader-mcp (npx)

Server: pdf-reader-mcp v2.1.0 by SylphxAI Transport: stdio via npx -y @sylphx/pdf-reader-mcp Auth: none — credential-free, zero config Processing: local (pdf.js-based, no external API calls for parsing)

Tool: read_pdf

Single tool with rich options:

MCP handshake

→ {"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"agent","version":"1.0.0"}}}
← {"jsonrpc":"2.0","id":1,"result":{"protocolVersion":"2025-03-26","capabilities":{"tools":{"listChanged":true},"logging":{},"completions":{}},"serverInfo":{"name":"pdf-reader-mcp","version":"2.1.0"},"instructions":"MCP Server for reading PDF files and extracting text, metadata, images, and page information."}}
→ {"jsonrpc":"2.0","method":"notifications/initialized"}

Test 1: Local PDF — full text + metadata + page count

→ {"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"read_pdf","arguments":{"sources":[{"path":"/tmp/test-pathfinder.pdf"}],"include_full_text":true,"include_metadata":true,"include_page_count":true,"include_tables":true}}}

← (two content blocks)
Block 1 (structured JSON):
{
  "results": [{
    "source": "/tmp/test-pathfinder.pdf",
    "success": true,
    "data": {
      "num_pages": 1,
      "info": {
        "PDFFormatVersion": "1.3",
        "IsLinearized": false,
        "IsAcroFormPresent": false,
        "CreationDate": "D:20260611052226Z"
      },
      "full_text": "Pathfinder Test Document\nThis is a test PDF..."
    }
  }]
}

Block 2 (page-by-page text):
[Page 1]
Pathfinder Test Document
This is a test PDF for verifying the @sylphx/pdf-reader-mcp server.
Page 1 of 1 - generated 2026-06-11.
Section: Authentication Flow
1. User submits credentials via POST /login
2. Server validates against the user database
3. On success, returns a signed JWT token
4. Client stores token in httpOnly cookie

Test 2: URL source — specific page extraction with metadata

→ {"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"read_pdf","arguments":{"sources":[{"url":"https://raw.githubusercontent.com/mozilla/pdf.js/master/test/pdfs/basicapi.pdf","pages":[1]}],"include_full_text":true,"include_page_count":true}}}

← {
  "results": [{
    "source": "https://raw.githubusercontent.com/mozilla/pdf.js/master/test/pdfs/basicapi.pdf",
    "success": true,
    "data": {
      "num_pages": 3,
      "info": {
        "PDFFormatVersion": "1.7",
        "Language": "en",
        "Title": "Basic API Test",
        "Author": "Brendan Dahl",
        "Creator": "pdf.js",
        "Producer": "TCPDF 5.9.133",
        "CreationDate": "D:20120410143026-07'00'"
      },
      "page_texts": [{"page": 1, "text": "Table Of Content..."}]
    }
  }]
}

Gotchas

• URL redirects: the server fails on URLs that issue HTTP redirects without a Location header (tested: W3C URLs return isError: true with "redirect without Location header"). Use direct download URLs.
• Multi-source batching: the sources array accepts multiple PDFs in one call — useful for batch processing.
• Page selection: use pages array for specific pages (1-indexed), omit for all pages.
• Two content blocks: responses come as two content items — first is structured JSON with metadata/info, second is human-readable page-by-