UNPKG

@sylphlab/pdf-reader-mcp

Version:

An MCP server providing tools to read PDF files.

328 lines (327 loc) 14.5 kB
import { z } from 'zod'; import * as pdfjsLib from 'pdfjs-dist/legacy/build/pdf.mjs'; import fs from 'node:fs/promises'; import { resolvePath } from '../utils/pathUtils.js'; import { McpError, ErrorCode } from '@modelcontextprotocol/sdk/types.js'; // Helper to parse page range strings (e.g., "1-3,5,7-") // Helper to parse a single range part (e.g., "1-3", "5", "7-") const parseRangePart = (part, pages) => { const trimmedPart = part.trim(); if (trimmedPart.includes('-')) { const [startStr, endStr] = trimmedPart.split('-'); if (startStr === undefined) { // Basic check throw new Error(`Invalid page range format: ${trimmedPart}`); } const start = parseInt(startStr, 10); const end = endStr === '' || endStr === undefined ? Infinity : parseInt(endStr, 10); if (isNaN(start) || isNaN(end) || start <= 0 || start > end) { throw new Error(`Invalid page range values: ${trimmedPart}`); } // Add a reasonable upper limit to prevent infinite loops for open ranges const practicalEnd = Math.min(end, start + 10000); // Limit range parsing depth for (let i = start; i <= practicalEnd; i++) { pages.add(i); } if (end === Infinity && practicalEnd === start + 10000) { console.warn(`[PDF Reader MCP] Open-ended range starting at ${String(start)} was truncated at page ${String(practicalEnd)} during parsing.`); } } else { const page = parseInt(trimmedPart, 10); if (isNaN(page) || page <= 0) { throw new Error(`Invalid page number: ${trimmedPart}`); } pages.add(page); } }; // Parses the complete page range string (e.g., "1-3,5,7-") const parsePageRanges = (ranges) => { const pages = new Set(); const parts = ranges.split(','); for (const part of parts) { parseRangePart(part, pages); // Delegate parsing of each part } if (pages.size === 0) { throw new Error('Page range string resulted in zero valid pages.'); } return Array.from(pages).sort((a, b) => a - b); }; // --- Zod Schemas --- const pageSpecifierSchema = z.union([ z.array(z.number().int().positive()).min(1), // Array of positive integers z .string() .min(1) .refine((val) => /^[0-9,-]+$/.test(val.replace(/\s/g, '')), { // Allow spaces but test without them message: 'Page string must contain only numbers, commas, and hyphens.', }), ]); const PdfSourceSchema = z .object({ path: z.string().min(1).optional().describe('Relative path to the local PDF file.'), url: z.string().url().optional().describe('URL of the PDF file.'), pages: pageSpecifierSchema .optional() .describe("Extract text only from specific pages (1-based) or ranges for *this specific source*. If provided, 'include_full_text' for the entire request is ignored for this source."), }) .strict() .refine((data) => !!(data.path && !data.url) || !!(!data.path && data.url), { // Use boolean coercion instead of || for truthiness check if needed, though refine expects boolean message: "Each source must have either 'path' or 'url', but not both.", }); const ReadPdfArgsSchema = z .object({ sources: z .array(PdfSourceSchema) .min(1) .describe('An array of PDF sources to process, each can optionally specify pages.'), include_full_text: z .boolean() .optional() .default(false) .describe("Include the full text content of each PDF (only if 'pages' is not specified for that source)."), include_metadata: z .boolean() .optional() .default(true) .describe('Include metadata and info objects for each PDF.'), include_page_count: z .boolean() .optional() .default(true) .describe('Include the total number of pages for each PDF.'), }) .strict(); // --- Helper Functions --- // Parses the page specification for a single source const getTargetPages = (sourcePages, sourceDescription) => { if (!sourcePages) { return undefined; } try { let targetPages; if (typeof sourcePages === 'string') { targetPages = parsePageRanges(sourcePages); } else { // Ensure array elements are positive integers if (sourcePages.some((p) => !Number.isInteger(p) || p <= 0)) { throw new Error('Page numbers in array must be positive integers.'); } targetPages = [...new Set(sourcePages)].sort((a, b) => a - b); } if (targetPages.length === 0) { // Check after potential Set deduplication throw new Error('Page specification resulted in an empty set of pages.'); } return targetPages; } catch (error) { const message = error instanceof Error ? error.message : String(error); // Throw McpError for invalid page specs caught during parsing throw new McpError(ErrorCode.InvalidParams, `Invalid page specification for source ${sourceDescription}: ${message}`); } }; // Loads the PDF document from path or URL const loadPdfDocument = async (source, // Explicitly allow undefined sourceDescription) => { let pdfDataSource; try { if (source.path) { const safePath = resolvePath(source.path); // resolvePath handles security checks pdfDataSource = await fs.readFile(safePath); } else if (source.url) { pdfDataSource = { url: source.url }; } else { // This case should be caught by Zod, but added for robustness throw new McpError(ErrorCode.InvalidParams, `Source ${sourceDescription} missing 'path' or 'url'.`); } } catch (err) { // Handle errors during path resolution or file reading let errorMessage; // Declare errorMessage here const message = err instanceof Error ? err.message : String(err); const errorCode = ErrorCode.InvalidRequest; // Default error code if (typeof err === 'object' && err !== null && 'code' in err && err.code === 'ENOENT' && source.path) { // Specific handling for file not found errorMessage = `File not found at '${source.path}'.`; // Optionally keep errorCode as InvalidRequest or change if needed } else { // Generic error for other file prep issues or resolvePath errors errorMessage = `Failed to prepare PDF source ${sourceDescription}. Reason: ${message}`; } throw new McpError(errorCode, errorMessage, { cause: err instanceof Error ? err : undefined }); } const loadingTask = pdfjsLib.getDocument(pdfDataSource); try { return await loadingTask.promise; } catch (err) { console.error(`[PDF Reader MCP] PDF.js loading error for ${sourceDescription}:`, err); const message = err instanceof Error ? err.message : String(err); // Use ?? for default message throw new McpError(ErrorCode.InvalidRequest, `Failed to load PDF document from ${sourceDescription}. Reason: ${message || 'Unknown loading error'}`, // Revert to || as message is likely always string here { cause: err instanceof Error ? err : undefined }); } }; // Extracts metadata and page count const extractMetadataAndPageCount = async (pdfDocument, includeMetadata, includePageCount) => { const output = {}; if (includePageCount) { output.num_pages = pdfDocument.numPages; } if (includeMetadata) { try { const pdfMetadata = await pdfDocument.getMetadata(); const infoData = pdfMetadata.info; if (infoData !== undefined) { output.info = infoData; } const metadataObj = pdfMetadata.metadata; const metadataData = metadataObj.getAll(); if (metadataData !== undefined) { output.metadata = metadataData; } } catch (metaError) { console.warn(`[PDF Reader MCP] Error extracting metadata: ${metaError instanceof Error ? metaError.message : String(metaError)}`); // Optionally add a warning to the result if metadata extraction fails partially } } return output; }; // Extracts text from specified pages const extractPageTexts = async (pdfDocument, pagesToProcess, sourceDescription) => { const extractedPageTexts = []; for (const pageNum of pagesToProcess) { let pageText = ''; try { const page = await pdfDocument.getPage(pageNum); const textContent = await page.getTextContent(); pageText = textContent.items .map((item) => item.str) // Type assertion .join(''); } catch (pageError) { const message = pageError instanceof Error ? pageError.message : String(pageError); console.warn(`[PDF Reader MCP] Error getting text content for page ${String(pageNum)} in ${sourceDescription}: ${message}` // Explicit string conversion ); pageText = `Error processing page: ${message}`; // Include error in text } extractedPageTexts.push({ page: pageNum, text: pageText }); } // Sorting is likely unnecessary if pagesToProcess was sorted, but keep for safety extractedPageTexts.sort((a, b) => a.page - b.page); return extractedPageTexts; }; // Determines the actual list of pages to process based on target pages and total pages const determinePagesToProcess = (targetPages, totalPages, includeFullText) => { let pagesToProcess = []; let invalidPages = []; if (targetPages) { // Filter target pages based on actual total pages pagesToProcess = targetPages.filter((p) => p <= totalPages); invalidPages = targetPages.filter((p) => p > totalPages); } else if (includeFullText) { // If no specific pages requested for this source, use global flag pagesToProcess = Array.from({ length: totalPages }, (_, i) => i + 1); } return { pagesToProcess, invalidPages }; }; // Processes a single PDF source const processSingleSource = async (source, globalIncludeFullText, globalIncludeMetadata, globalIncludePageCount) => { const sourceDescription = source.path ?? source.url ?? 'unknown source'; let individualResult = { source: sourceDescription, success: false }; try { // 1. Parse target pages for this source (throws McpError on invalid spec) const targetPages = getTargetPages(source.pages, sourceDescription); // 2. Load PDF Document (throws McpError on loading failure) // Destructure to remove 'pages' before passing to loadPdfDocument due to exactOptionalPropertyTypes const { pages: _pages, ...loadArgs } = source; const pdfDocument = await loadPdfDocument(loadArgs, sourceDescription); const totalPages = pdfDocument.numPages; // 3. Extract Metadata & Page Count const metadataOutput = await extractMetadataAndPageCount(pdfDocument, globalIncludeMetadata, globalIncludePageCount); const output = { ...metadataOutput }; // Start building output // 4. Determine actual pages to process const { pagesToProcess, invalidPages } = determinePagesToProcess(targetPages, totalPages, globalIncludeFullText // Pass the global flag ); // Add warnings for invalid requested pages if (invalidPages.length > 0) { output.warnings = output.warnings ?? []; output.warnings.push(`Requested page numbers ${invalidPages.join(', ')} exceed total pages (${String(totalPages)}).`); } // 5. Extract Text (if needed) if (pagesToProcess.length > 0) { const extractedPageTexts = await extractPageTexts(pdfDocument, pagesToProcess, sourceDescription); if (targetPages) { // If specific pages were requested for *this source* output.page_texts = extractedPageTexts; } else { // Only assign full_text if pages were NOT specified for this source output.full_text = extractedPageTexts.map((p) => p.text).join('\n\n'); } } individualResult = { ...individualResult, data: output, success: true }; } catch (error) { let errorMessage = `Failed to process PDF from ${sourceDescription}.`; if (error instanceof McpError) { errorMessage = error.message; // Use message from McpError directly } else if (error instanceof Error) { errorMessage += ` Reason: ${error.message}`; } else { errorMessage += ` Unknown error: ${JSON.stringify(error)}`; } individualResult.error = errorMessage; individualResult.success = false; delete individualResult.data; // Ensure no data on error } return individualResult; }; // --- Main Handler Function --- export const handleReadPdfFunc = async (args) => { let parsedArgs; try { parsedArgs = ReadPdfArgsSchema.parse(args); } catch (error) { if (error instanceof z.ZodError) { throw new McpError(ErrorCode.InvalidParams, `Invalid arguments: ${error.errors.map((e) => `${e.path.join('.')} (${e.message})`).join(', ')}`); } // Added fallback for non-Zod errors during parsing const message = error instanceof Error ? error.message : String(error); throw new McpError(ErrorCode.InvalidParams, `Argument validation failed: ${message}`); } const { sources, include_full_text, include_metadata, include_page_count } = parsedArgs; // Process all sources concurrently const results = await Promise.all(sources.map((source) => processSingleSource(source, include_full_text, include_metadata, include_page_count))); return { content: [ { type: 'text', text: JSON.stringify({ results }, null, 2), }, ], }; }; // Export the consolidated ToolDefinition export const readPdfToolDefinition = { name: 'read_pdf', description: 'Reads content/metadata from one or more PDFs (local/URL). Each source can specify pages to extract.', schema: ReadPdfArgsSchema, handler: handleReadPdfFunc, };