sad-mcp 0.1.13 → 0.1.15

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+        // No version file — treat as outdated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/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: {
@@ -64,13 +78,17 @@ 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: {
                         name: {
                             type: "string",
-                            description: "The file name (or partial name) to retrieve. Matched against file names from search_materials or list_materials results.",
+                            description: "The file name or path (or partial match) to retrieve. Use the path from search_materials results for exact matching (e.g., 'מבחנים-לסטודנטים/2024-א-א/מבחן.pdf').",
+                        },
+                        page: {
+                            type: "number",
+                            description: "Page number (1-indexed). Each page is ~5000 characters. Defaults to 1.",
                         },
                         user_question: {
                             type: "string",
@@ -142,6 +160,7 @@ export function registerToolHandlers(server) {
                 if (matches.length > 0 || nameMatch) {
                     results.push({
                         fileName: file.name,
+                        path: file.path,
                         category: categorizeFile(file),
                         matchCount: nameMatch ? matches.length + 100 : matches.length, // Boost file-name matches
                         preview: matches.length > 0
@@ -158,6 +177,7 @@ export function registerToolHandlers(server) {
                 if (file.name.toLowerCase().includes(queryLower) || file.path.toLowerCase().includes(queryLower)) {
                     results.push({
                         fileName: file.name,
+                        path: file.path,
                         category: categorizeFile(file),
                         matchCount: 100,
                         preview: `(file name matches "${query}" — use get_material to read)`,
@@ -168,7 +188,7 @@ export function registerToolHandlers(server) {
             results.sort((a, b) => b.matchCount - a.matchCount);
             const responseText = results.length === 0
                 ? `No results found for "${query}" in course materials.`
-                : `Found "${query}" in ${results.length} file(s). Use get_material to read the most relevant one(s):\n\n${results.map((r) => `- ${r.fileName} [${r.category}] (${r.matchCount} matches) — "${r.preview}"`).join("\n")}`;
+                : `Found "${query}" in ${results.length} file(s). Use get_material with the file path to read the most relevant one(s):\n\n${results.map((r) => `- ${r.path} [${r.category}] (${r.matchCount} matches) — "${r.preview}"`).join("\n")}`;
             trackToolCall(name, toolArgs, { resultCount: results.length, success: results.length > 0, responseChars: responseText.length }, Date.now() - startTime);
             return { content: [{ type: "text", text: responseText }] };
         }
@@ -181,18 +201,18 @@ export function registerToolHandlers(server) {
             }
             await ensureTextCache();
             const queryLower = queryName.toLowerCase();
-            // First: check text cache
+            // First: check text cache (match against both name and path)
             let bestMatch = null;
             for (const [, entry] of textCache) {
-                if (entry.file.name.toLowerCase().includes(queryLower)) {
+                if (entry.file.name.toLowerCase().includes(queryLower) || entry.file.path.toLowerCase().includes(queryLower)) {
                     bestMatch = entry;
                     break;
                 }
             }
-            // Fallback: search all files by name and attempt fresh extraction
+            // Fallback: search all files by name/path and attempt fresh extraction
             if (!bestMatch) {
                 const allFiles = await listAllFiles();
-                const matchedFile = allFiles.find(f => f.name.toLowerCase().includes(queryLower));
+                const matchedFile = allFiles.find(f => f.name.toLowerCase().includes(queryLower) || f.path.toLowerCase().includes(queryLower));
                 if (matchedFile && isExtractable(matchedFile)) {
                     try {
                         const buffer = await downloadFile(matchedFile);
@@ -215,12 +235,19 @@ 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/package.json

@@ -1,6 +1,6 @@
 {
   "name": "sad-mcp",
-  "version": "0.1.13",
+  "version": "0.1.15",
   "description": "MCP server for Software Analysis and Design course materials at BGU",
   "type": "module",
   "bin": {