@cachly-dev/mcp-server 0.5.48 → 0.5.49

package/dist/index.d.ts

@@ -33,8 +33,9 @@ declare function expandCrossLingual(token: string): string[];
  * Features:
  *  • CJK → character bigrams
  *  • Katakana → additionally emits Hepburn romaji tokens (enables romaji queries)
- *  • Arabic/Hebrew → word tokenization with light Arabic stemming
- *  • All scripts → cross-lingual synonym expansion (EN↔JA↔ZH↔KO↔AR↔HE)
+ *  • Arabic → word tokenization + iterative prefix stemming (al-, wa-, bi-, li-, fa-)
+ *  • Hebrew → word tokenization + iterative prefix stemming (ha-, ve-, be-, le-, mi-, ke-)
+ *  • All scripts → cross-lingual synonym expansion (EN↔JA↔ZH↔KO↔AR↔HE↔DE↔FR↔ES↔IT↔PT)
  */
 declare function tokenize(text: string): string[];
 /**

package/dist/index.js

@@ -47,7 +47,7 @@ import { Redis } from 'ioredis';
 const API_URL = process.env.CACHLY_API_URL ?? 'https://api.cachly.dev';
 const JWT = process.env.CACHLY_JWT ?? '';
 const EMBED_MODEL = process.env.CACHLY_EMBED_MODEL ?? '';
-const CURRENT_VERSION = '0.5.48';
+const CURRENT_VERSION = '0.5.49';
 /**
  * ┌──────────────────────────────────────────────────────────────────────┐
  * │  EMBEDDING PROVIDER — pluggable, client-side first                  │
@@ -409,6 +409,10 @@ const CJK_RE = /[\u3040-\u30ff\u3400-\u4dbf\u4e00-\u9fff\uf900-\ufaff\uac00-\ud7
 const SEGMENT_RE = /([\u3040-\u30ff\u3400-\u4dbf\u4e00-\u9fff\uf900-\ufaff\uac00-\ud7af]+|[^\u3040-\u30ff\u3400-\u4dbf\u4e00-\u9fff\uf900-\ufaff\uac00-\ud7af]+)/g;
 // RTL Unicode ranges: Arabic (U+0600–U+06FF + extended), Hebrew (U+0590–U+05FF)
 const RTL_RE = /[\u0590-\u05ff\u0600-\u06ff\u0750-\u077f\u08a0-\u08ff\ufb1d-\ufb4f\ufb50-\ufdff\ufe70-\ufeff]/;
+// Distinguish Arabic from Hebrew within RTL segments
+const HEBREW_CHAR_RE = /[\u0590-\u05ff\ufb1d-\ufb4f]/;
+const ARABIC_CHAR_RE = /[\u0600-\u06ff\u0750-\u077f\u08a0-\u08ff\ufb50-\ufdff\ufe70-\ufeff]/;
+;
 // ── Katakana → Romaji conversion table (Hepburn system) ──────────────────────
 // Digraphs must be listed before single chars so they match first.
 const KATA_ROMAJI_TABLE = [
@@ -509,6 +513,33 @@ function arabicLightStem(word) {
     }
     return result;
 }
+/**
+ * Hebrew light stemmer — strips the definite article and common prefix particles
+ * that attach directly to words (no space) in Hebrew.
+ *
+ * Handles:
+ *   ה (ha-) — definite article:   הפריסה → פריסה
+ *   ו (ve-/u-) — conjunction “and”: ופריסה → פריסה
+ *   ב (be-/bi-) — preposition “in”: בסביבה → סביבה
+ *   ל (le-/li-) — preposition “for”: לשרת → שרת
+ *   מ (mi-/me-) — preposition “from”: מהשרת → השרת → שרת
+ *   כ (ke-/ki-) — preposition “like”: כשרת → שרת
+ *
+ * Iterative (max 3 passes): מהפריסה → הפריסה → פריסה.
+ * Only strips when result is still ≥ 3 chars.
+ */
+function hebrewLightStem(word) {
+    let result = word;
+    for (let i = 0; i < 3; i++) {
+        const prev = result;
+        if (result.length > 3 && 'הובלמכ'.includes(result[0]) && HEBREW_CHAR_RE.test(result[1])) {
+            result = result.slice(1);
+        }
+        if (result === prev)
+            break;
+    }
+    return result;
+}
 // ── Cross-lingual tech term dictionary ────────────────────────────────────────
 // Maps tokens (English, Japanese katakana, Chinese, Korean, Arabic, Hebrew)
 // to their equivalents in other languages. Used to expand tokens at search time
@@ -517,8 +548,8 @@ function arabicLightStem(word) {
 // Keys must be lowercase. CJK values will be bigram-expanded at use time.
 const CROSS_LINGUAL_MAP = new Map([
     // English → all
-    ['deploy', ['デプロイ', '部署', '배포', 'deployment', 'deploying', 'نشر', 'bereitstellen', 'bereitstellung', 'déployer', 'déploiement', 'desplegar', 'despliegue', 'distribuire', 'distribuzione', 'implantar', 'implantação']],
-    ['deployment', ['デプロイ', '部署', '배포', 'deploy', 'bereitstellung', 'déploiement', 'despliegue', 'distribuzione', 'implantação']],
+    ['deploy', ['デプロイ', '部署', '배포', 'deployment', 'deploying', 'نشر', 'פריסה', 'bereitstellen', 'bereitstellung', 'déployer', 'déploiement', 'desplegar', 'despliegue', 'distribuire', 'distribuzione', 'implantar', 'implantação']],
+    ['deployment', ['デプロイ', '部署', '배포', 'deploy', 'نشر', 'פריסה', 'bereitstellung', 'déploiement', 'despliegue', 'distribuzione', 'implantação']],
     ['container', ['コンテナ', '容器', '컨테이너', 'docker']],
     ['server', ['サーバー', 'サーバ', '服务器', '서버', 'سيرفر', 'שרת']],
     ['database', ['データベース', '数据库', '데이터베이스', 'db', 'قاعدة البيانات', 'datenbank', 'base de données', 'base de datos', 'banco de dados']],
@@ -530,6 +561,9 @@ const CROSS_LINGUAL_MAP = new Map([
     ['test', ['テスト', '测试', '테스트', 'اختبار', 'בדיקה', 'testen', 'tester', 'probar', 'testare', 'testar']],
     ['auth', ['認証', '认证', '인증', 'authentication', 'login', 'oauth', 'مصادقة', 'אימות', 'authentifizierung', 'authentification', 'autenticación', 'autenticazione', 'autenticação']],
     ['authentication', ['auth', 'مصادقة', 'אימות', '認証', '认证', '인증', 'login', 'oauth']],
+    ['environment', ['環境', '环境', '환경', 'env', 'بيئة', 'סביבה', 'umgebung', 'environnement', 'entorno', 'ambiente']],
+    ['secret', ['key', 'token', 'password', 'jwt', 'مفتاح', 'מפתח', 'geheimnis', 'schlüssel', 'clave', 'clé', 'chiave', 'chave']],
+    ['problem', ['error', 'issue', 'bug', 'failure', 'مشكلة', 'بعيا', '問題', '問題', '오류', 'fehler', 'problème', 'problema']],
     ['kubernetes', ['クベルネテス', 'k8s', 'kube']],
     ['network', ['ネットワーク', '网络', '네트워크', 'شبكة', 'רשת', 'netzwerk', 'réseau', 'red', 'rete', 'rede']],
     ['timeout', ['タイムアウト', '超时', '타임아웃', 'مهلة']],
@@ -707,6 +741,13 @@ const CROSS_LINGUAL_MAP = new Map([
     ['قرص', ['disk', 'storage']],
     ['خادم', ['server']],
     ['برنامج', ['application', 'software', 'app']],
+    // Arabic terms from blog examples
+    ['مشكلة', ['error', 'issue', 'bug', 'problem', 'failure']],
+    ['بيئة', ['environment', 'env']],
+    ['متغيرات', ['variables', 'env', 'environment']],
+    ['إنتاج', ['production', 'prod']],
+    ['سري', ['secret', 'key', 'token']],
+    ['مفتاح سري', ['secret', 'key']],
     // Hebrew → English
     ['שגיאה', ['error']],
     ['תיקון', ['fix']],
@@ -742,6 +783,13 @@ const CROSS_LINGUAL_MAP = new Map([
     ['ממשק תכנות', ['api', 'interface']],
     ['ביצועים', ['performance']],
     ['אחסון', ['storage', 'disk']],
+    // Hebrew terms from blog examples
+    ['פריסה', ['deploy', 'deployment']],
+    ['בעיה', ['error', 'issue', 'bug', 'problem']],
+    ['סביבה', ['environment', 'env']],
+    ['ייצור', ['production', 'prod']],
+    ['סודי', ['secret', 'key', 'token']],
+    ['מפתח סודי', ['secret', 'key']],
     // ── European languages → English ─────────────────────────────────────────
     // German (Deutsch) → English
     ['fehler', ['error', 'err']],
@@ -939,8 +987,9 @@ function expandCrossLingual(token) {
  * Features:
  *  • CJK → character bigrams
  *  • Katakana → additionally emits Hepburn romaji tokens (enables romaji queries)
- *  • Arabic/Hebrew → word tokenization with light Arabic stemming
- *  • All scripts → cross-lingual synonym expansion (EN↔JA↔ZH↔KO↔AR↔HE)
+ *  • Arabic → word tokenization + iterative prefix stemming (al-, wa-, bi-, li-, fa-)
+ *  • Hebrew → word tokenization + iterative prefix stemming (ha-, ve-, be-, le-, mi-, ke-)
+ *  • All scripts → cross-lingual synonym expansion (EN↔JA↔ZH↔KO↔AR↔HE↔DE↔FR↔ES↔IT↔PT)
  */
 function tokenize(text) {
     const tokens = [];
@@ -979,12 +1028,15 @@ function tokenize(text) {
             }
         }
         else if (RTL_RE.test(seg)) {
-            // RTL segment (Arabic / Hebrew) — space-separated words with light stemming
+            // RTL segment — split on spaces, apply language-specific prefix stemming
+            // Arabic: strips ال/و/ب/ل/ف/ك prefixes iteratively
+            // Hebrew: strips ה/ו/ב/ל/מ/כ prefixes iteratively
+            const isHebrew = HEBREW_CHAR_RE.test(seg);
             const words = seg.trim().split(/\s+/);
             for (const w of words) {
                 if (w.length <= 1 || STOPWORDS.has(w))
                     continue;
-                const stemmed = arabicLightStem(w);
+                const stemmed = isHebrew ? hebrewLightStem(w) : arabicLightStem(w);
                 if (stemmed.length > 1 && !STOPWORDS.has(stemmed))
                     tokens.push(stemmed);
             }