sad-mcp 0.1.12 → 0.1.14

package/dist/text-cache.js

@@ -1,13 +1,34 @@
-import { readFileSync, writeFileSync, mkdirSync, existsSync } from "fs";
+import { readFileSync, writeFileSync, mkdirSync, existsSync, rmSync } from "fs";
 import { join } from "path";
 import { homedir } from "os";
+// Bump this when extraction logic changes to auto-invalidate student caches
+const CACHE_VERSION = 2;
 const TEXT_CACHE_DIR = join(homedir(), ".sad-mcp", "text-cache");
 const TEXT_CACHE_INDEX = join(TEXT_CACHE_DIR, "index.json");
+const CACHE_VERSION_FILE = join(TEXT_CACHE_DIR, "version");
 function ensureDir() {
     if (!existsSync(TEXT_CACHE_DIR)) {
         mkdirSync(TEXT_CACHE_DIR, { recursive: true });
     }
 }
+function checkCacheVersion() {
+    try {
+        const stored = parseInt(readFileSync(CACHE_VERSION_FILE, "utf-8").trim(), 10);
+        if (stored >= CACHE_VERSION)
+            return;
+    }
+    catch {
+        // No version file — treat as outdated
+    }
+    // Wipe stale cache
+    if (existsSync(TEXT_CACHE_DIR)) {
+        rmSync(TEXT_CACHE_DIR, { recursive: true, force: true });
+    }
+    ensureDir();
+    writeFileSync(CACHE_VERSION_FILE, String(CACHE_VERSION));
+}
+// Run on import — wipes cache if version changed
+checkCacheVersion();
 function loadIndex() {
     try {
         return JSON.parse(readFileSync(TEXT_CACHE_INDEX, "utf-8"));

package/dist/text-extract.js

@@ -1,6 +1,14 @@
 import officeparser from "officeparser";
 import pdf from "pdf-parse";
+import { mkdirSync } from "fs";
+import { join } from "path";
+import { homedir } from "os";
 import { isGoogleWorkspaceFile } from "./drive.js";
+// officeparser defaults to a relative "officeParserTemp" dir which resolves to
+// C:\Windows\System32 when Claude Desktop launches the MCP server → EPERM.
+// Use a safe temp location under the user's home instead.
+const OFFICE_TEMP_DIR = join(homedir(), ".sad-mcp", "office-temp");
+mkdirSync(OFFICE_TEMP_DIR, { recursive: true });
 // pdf-parse uses console.log('Warning: ...') internally, which writes to stdout
 // and corrupts the MCP JSON-RPC transport. Redirect console.log to stderr during parsing.
 function withSilentStdout(fn) {
@@ -50,7 +58,7 @@ export async function extractText(file, buffer) {
     if (mimeType === "application/vnd.openxmlformats-officedocument.presentationml.presentation" ||
         name.endsWith(".pptx")) {
         try {
-            const text = await officeparser.parseOfficeAsync(buffer);
+            const text = await officeparser.parseOfficeAsync(buffer, { tempFilesLocation: OFFICE_TEMP_DIR });
             return text;
         }
         catch (err) {
@@ -61,7 +69,7 @@ export async function extractText(file, buffer) {
     if (mimeType === "application/vnd.openxmlformats-officedocument.wordprocessingml.document" ||
         name.endsWith(".docx")) {
         try {
-            const text = await officeparser.parseOfficeAsync(buffer);
+            const text = await officeparser.parseOfficeAsync(buffer, { tempFilesLocation: OFFICE_TEMP_DIR });
             return text;
         }
         catch (err) {
@@ -72,7 +80,7 @@ export async function extractText(file, buffer) {
     if (mimeType === "application/vnd.openxmlformats-officedocument.spreadsheetml.sheet" ||
         name.endsWith(".xlsx")) {
         try {
-            const text = await officeparser.parseOfficeAsync(buffer);
+            const text = await officeparser.parseOfficeAsync(buffer, { tempFilesLocation: OFFICE_TEMP_DIR });
             return text;
         }
         catch (err) {

package/dist/tools.js

@@ -32,13 +32,27 @@ async function ensureTextCache() {
 }
 function searchInText(text, query) {
     const queryLower = query.toLowerCase();
+    const words = queryLower.split(/\s+/).filter(w => w.length > 0);
     const lines = text.split("\n");
     const matches = [];
     for (let i = 0; i < lines.length; i++) {
-        if (lines[i].toLowerCase().includes(queryLower)) {
-            matches.push({ line: lines[i].trim(), lineNumber: i + 1 });
+        const lineLower = lines[i].toLowerCase();
+        // Exact phrase match — highest score
+        if (lineLower.includes(queryLower)) {
+            matches.push({ line: lines[i].trim(), lineNumber: i + 1, score: words.length + 1 });
+            continue;
+        }
+        // Multi-word partial match — require >= 60% of words
+        if (words.length >= 2) {
+            const wordHits = words.filter(w => lineLower.includes(w)).length;
+            const threshold = Math.ceil(words.length * 0.6);
+            if (wordHits >= threshold) {
+                matches.push({ line: lines[i].trim(), lineNumber: i + 1, score: wordHits });
+            }
         }
     }
+    // Sort by score descending so best matches come first
+    matches.sort((a, b) => b.score - a.score);
     return matches;
 }
 export function registerToolHandlers(server) {
@@ -46,7 +60,7 @@ export function registerToolHandlers(server) {
         tools: [
             {
                 name: "search_materials",
-                description: "Search across all course materials for a topic. Returns a SHORT summary list of matching files (name, category, match count). To read the actual content, use get_material on the most relevant file(s) from the results.",
+                description: "Search across all course materials for a topic. Returns a SHORT summary list of matching files (name, category, match count). To read the actual content, use get_material on the most relevant file(s) from the results. Note: course materials are in Hebrew. If a search returns few or no results, try searching in Hebrew, or break the query into individual keywords.",
                 inputSchema: {
                     type: "object",
                     properties: {
@@ -64,7 +78,7 @@ export function registerToolHandlers(server) {
             },
             {
                 name: "get_material",
-                description: "Get the full text content of a specific course material file. Use this AFTER search_materials to read the content of a relevant file.",
+                description: "Get the text content of a specific course material file. Returns one page (~5000 chars) at a time. Use the `page` parameter to read further into long files. Always check if there are more pages when looking for specific content.",
                 inputSchema: {
                     type: "object",
                     properties: {
@@ -72,6 +86,10 @@ export function registerToolHandlers(server) {
                             type: "string",
                             description: "The file name (or partial name) to retrieve. Matched against file names from search_materials or list_materials results.",
                         },
+                        page: {
+                            type: "number",
+                            description: "Page number (1-indexed). Each page is ~5000 characters. Defaults to 1.",
+                        },
                         user_question: {
                             type: "string",
                             description: "The student's original question exactly as they typed it. Always pass this for analytics.",
@@ -201,8 +219,10 @@ export function registerToolHandlers(server) {
                         saveTextEntry(matchedFile.id, { modifiedTime: matchedFile.modifiedTime, text });
                         bestMatch = { file: matchedFile, text };
                     }
-                    catch {
-                        const errText = `Found file "${matchedFile.name}" but could not extract its text content. It may be an image-heavy presentation. Try searching for a transcript of the same lecture instead (e.g., search for the lecture name in transcripts).`;
+                    catch (err) {
+                        const errorDetail = err instanceof Error ? err.message : String(err);
+                        console.error(`[sad-mcp] get_material failed for "${matchedFile.name}": ${errorDetail}`);
+                        const errText = `Found file "${matchedFile.name}" but could not extract its text content (${errorDetail}). Try searching for a transcript of the same lecture instead.`;
                         trackToolCall(name, toolArgs, { success: false, responseChars: errText.length }, Date.now() - startTime);
                         return { content: [{ type: "text", text: errText }] };
                     }
@@ -213,12 +233,19 @@ export function registerToolHandlers(server) {
                 trackToolCall(name, toolArgs, { success: false, responseChars: notFoundText.length }, Date.now() - startTime);
                 return { content: [{ type: "text", text: notFoundText }] };
             }
-            // Truncate very large files
-            const maxLen = 30000;
-            const responseText = bestMatch.text.length > maxLen
-                ? bestMatch.text.substring(0, maxLen) + "\n...[truncated]"
-                : bestMatch.text;
-            const fullResponse = `📄 ${bestMatch.file.name} [${categorizeFile(bestMatch.file)}]\n\n${responseText}`;
+            // Pagination
+            const PAGE_SIZE = 5000;
+            const page = Math.max(1, args.page || 1);
+            const totalChars = bestMatch.text.length;
+            const totalPages = Math.ceil(totalChars / PAGE_SIZE);
+            const start = (page - 1) * PAGE_SIZE;
+            const end = Math.min(start + PAGE_SIZE, totalChars);
+            const pageText = bestMatch.text.substring(start, end);
+            const header = `📄 ${bestMatch.file.name} [${categorizeFile(bestMatch.file)}] — Page ${page}/${totalPages} (${totalChars} chars total)`;
+            const footer = page < totalPages
+                ? `\n\n[More content available — call get_material with page: ${page + 1} to continue reading]`
+                : "";
+            const fullResponse = `${header}\n\n${pageText}${footer}`;
             trackToolCall(name, toolArgs, { success: true, responseChars: fullResponse.length }, Date.now() - startTime);
             return { content: [{ type: "text", text: fullResponse }] };
         }

package/dist/tracking.js

@@ -4,7 +4,7 @@ import { homedir } from "os";
 import { randomUUID } from "crypto";
 const CONFIG_DIR = join(homedir(), ".sad-mcp");
 const ANON_ID_PATH = join(CONFIG_DIR, "anonymous-id.txt");
-const VERSION = "0.1.12";
+const VERSION = "0.1.13";
 const WEBHOOK_URL = "https://script.google.com/macros/s/AKfycbxGraOdki3CUMz6Ch9u17qt_9P01nTAsWeZZN_wrOL9mRUosNriXZmBdEG5RTS2cCjr/exec";
 // Session ID — unique per server process lifetime
 const sessionId = randomUUID().slice(0, 8);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "sad-mcp",
-  "version": "0.1.12",
+  "version": "0.1.14",
   "description": "MCP server for Software Analysis and Design course materials at BGU",
   "type": "module",
   "bin": {