PDF Reader MCP Server

Instructions

Implementation Reference

• src/handlers/readPdf.ts:145-225 (handler)

  Main implementation of the read_pdf tool handler. Orchestrates PDF loading, page processing, text/image extraction, and returns structured content using MCP SDK.
  export const readPdf = tool() .description( 'Reads content/metadata/images from one or more PDFs (local/URL). Each source can specify pages to extract.' ) .input(readPdfArgsSchema) .handler(async ({ input }) => { const { sources, include_full_text, include_metadata, include_page_count, include_images } = input; // Process sources with concurrency limit to prevent memory exhaustion // Processing large PDFs concurrently can consume significant memory const MAX_CONCURRENT_SOURCES = 3; const results: PdfSourceResult[] = []; const options = { includeFullText: include_full_text ?? false, includeMetadata: include_metadata ?? true, includePageCount: include_page_count ?? true, includeImages: include_images ?? false, }; for (let i = 0; i < sources.length; i += MAX_CONCURRENT_SOURCES) { const batch = sources.slice(i, i + MAX_CONCURRENT_SOURCES); const batchResults = await Promise.all( batch.map((source) => processSingleSource(source, options)) ); results.push(...batchResults); } // Check if all sources failed const allFailed = results.every((r) => !r.success); if (allFailed) { const errorMessages = results.map((r) => r.error).join('; '); return toolError(`All PDF sources failed to process: ${errorMessages}`); } // Build content parts - start with structured JSON for backward compatibility const content: Array<ReturnType<typeof text> | ReturnType<typeof image>> = []; // Strip image data and page_contents from JSON to keep it manageable const resultsForJson = results.map((result) => { if (result.data) { const { images, page_contents, ...dataWithoutBinaryContent } = result.data; // Include image count and metadata in JSON, but not the base64 data if (images) { const imageInfo = images.map((img) => ({ page: img.page, index: img.index, width: img.width, height: img.height, format: img.format, })); return { ...result, data: { ...dataWithoutBinaryContent, image_info: imageInfo } }; } return { ...result, data: dataWithoutBinaryContent }; } return result; }); // First content part: Structured JSON results content.push(text(JSON.stringify({ results: resultsForJson }, null, 2))); // Add page content in exact Y-coordinate order for (const result of results) { if (!result.success || !result.data?.page_contents) continue; // Process each page's content items in order for (const pageContent of result.data.page_contents) { for (const item of pageContent.items) { if (item.type === 'text' && item.textContent) { // Add text content part content.push(text(item.textContent)); } else if (item.type === 'image' && item.imageData) { // Add image content part (all images are now encoded as PNG) content.push(image(item.imageData.data, 'image/png')); } } } } return content; });
• src/schemas/readPdf.ts:32-50 (schema)

  Vex schema for validating input arguments to the read_pdf tool, including sources, page selections, and inclusion flags.
  export const readPdfArgsSchema = object({ sources: array(pdfSourceSchema), include_full_text: optional( bool( description( "Include the full text content of each PDF (only if 'pages' is not specified for that source)." ) ) ), include_metadata: optional(bool(description('Include metadata and info objects for each PDF.'))), include_page_count: optional( bool(description('Include the total number of pages for each PDF.')) ), include_images: optional( bool( description('Extract and include embedded images from the PDF pages as base64-encoded data.') ) ), });
• src/index.ts:4-13 (registration)

  Import and registration of the read_pdf tool in the MCP server creation.
  import { readPdf } from './handlers/readPdf.js'; const server = createServer({ name: 'pdf-reader-mcp', version: '1.3.0', instructions: 'MCP Server for reading PDF files and extracting text, metadata, images, and page information.', tools: { read_pdf: readPdf }, transport: stdio(), });