@cachly-dev/mcp-server 0.5.60 → 0.5.70

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 = '0.5.60';
+const CURRENT_VERSION = '0.5.70';
 /**
  * ┌──────────────────────────────────────────────────────────────────────┐
  * │  EMBEDDING PROVIDER — pluggable, client-side first                  │
@@ -437,6 +437,23 @@ const STOPWORDS = new Set([
     'çok', 'en', 'her', 'hiç', 'bazı', 'tüm', 'bütün', 'hem', 'veya',
     'bu', 'şu', 'ben', 'sen', 'biz', 'siz', 'onlar', 'benim', 'senin',
     'nasıl', 'neden', 'niçin', 'nerede', 'ne', 'hangi', 'kaç',
+    // ── Polish — common prepositions, pronouns, particles ─────────────────────
+    'w', 'na', 'z', 'do', 'się', 'że', 'to', 'jest', 'i', 'a', 'nie',
+    'tak', 'jak', 'czy', 'po', 'o', 'ale', 'go', 'mu', 'jej', 'ich',
+    'tego', 'tej', 'te', 'ten', 'ta', 'są', 'był', 'była', 'było', 'byli',
+    'będzie', 'będą', 'ma', 'mam', 'masz', 'mają', 'ze', 'co', 'już',
+    'przez', 'przy', 'za', 'bez', 'nad', 'pod', 'przed', 'po', 'między',
+    'kiedy', 'gdzie', 'dlaczego', 'który', 'która', 'które', 'tylko', 'też',
+    // ── Czech — common words ──────────────────────────────────────────────────
+    'v', 'na', 'z', 'do', 'se', 'že', 'to', 'je', 'a', 'ne', 'pro',
+    'ale', 'jak', 'by', 'byl', 'být', 'jsem', 'jsou', 'má', 'mám',
+    'ten', 'ta', 'ty', 'tato', 'jeho', 'její', 'jejich', 'také', 'jen',
+    'kde', 'kdy', 'proč', 'který', 'která', 'které', 'co', 'při', 'bez',
+    // ── Bengali — common particles, pronouns, verbs ───────────────────────────
+    'এই', 'এটি', 'এটা', 'এর', 'তার', 'তারা', 'আমি', 'তুমি', 'সে', 'আমরা',
+    'এবং', 'কিন্তু', 'বা', 'না', 'হ্যাঁ', 'যে', 'কি', 'কে', 'কী', 'থেকে',
+    'দিয়ে', 'জন্য', 'সাথে', 'মধ্যে', 'উপর', 'নিচে', 'আছে', 'ছিল', 'হয়',
+    'করা', 'করে', 'করেছে', 'হবে', 'হয়েছে', 'একটি', 'একটা', 'অনেক', 'সব',
 ]);
 // CJK Unicode ranges used for bigram extraction
 const CJK_RE = /[\u3040-\u30ff\u3400-\u4dbf\u4e00-\u9fff\uf900-\ufaff\uac00-\ud7af]/;
@@ -454,6 +471,8 @@ const DEVANAGARI_RE = /[\u0900-\u097f]/;
 const CYRILLIC_RE = /[\u0400-\u04ff]/;
 // Turkish uses Latin alphabet but has unique chars (ğ, ş, ı, ö, ü, ç) — detected by these
 const TURKISH_CHAR_RE = /[\u011f\u015f\u0131\u0130\u00e7]/; // ğ ş ı İ ç
+// Bengali (Bangla) — U+0980–U+09FF
+const BENGALI_RE = /[\u0980-\u09ff]/;
 // ── Katakana → Romaji conversion table (Hepburn system) ──────────────────────
 // Digraphs must be listed before single chars so they match first.
 const KATA_ROMAJI_TABLE = [
@@ -780,12 +799,134 @@ function turkishLightStem(word) {
         return w.slice(0, -1);
     return w;
 }
+/**
+ * Bengali (Bangla) light stemmer — strips common verbal and nominal suffixes.
+ * Bengali is an Indo-Aryan language with moderate morphological complexity.
+ * Only strips when result is still ≥ 2 chars.
+ *
+ * Covers:
+ *  করছে/করেছে → কর  (progressive/perfect aspect)
+ *  সার্ভারগুলো → সার্ভার  (-গুলো plural)
+ *  সমস্যাটি → সমস্যা  (-টি singular marker)
+ */
+function bengaliLightStem(word) {
+    let w = word;
+    // Plural / collective markers (longest first)
+    if (w.endsWith('\u0997\u09c1\u09b2\u09cb') && w.length > 5)
+        return w.slice(0, -4); // গুলো (-gulo plural)
+    if (w.endsWith('\u0997\u09c1\u09b2\u09bf') && w.length > 5)
+        return w.slice(0, -4); // গুলি (-guli plural)
+    if (w.endsWith('\u09a6\u09c7\u09b0') && w.length > 4)
+        return w.slice(0, -3); // দের (genitive plural)
+    if (w.endsWith('\u09a6\u09bf\u0997\u09c7') && w.length > 5)
+        return w.slice(0, -4); // দিগে
+    // Definiteness / case markers
+    if (w.endsWith('\u099f\u09bf') && w.length > 3)
+        return w.slice(0, -2); // টি (singular definite)
+    if (w.endsWith('\u099f\u09be') && w.length > 3)
+        return w.slice(0, -2); // টা (singular definite)
+    if (w.endsWith('\u0996\u09be\u09a8\u09be') && w.length > 5)
+        return w.slice(0, -4); // খানা
+    // Verbal suffixes
+    if (w.endsWith('\u099b\u09c7') && w.length > 3)
+        return w.slice(0, -2); // ছে (progressive)
+    if (w.endsWith('\u099b\u09bf\u09b2') && w.length > 4)
+        return w.slice(0, -3); // ছিল (past progressive)
+    if (w.endsWith('\u09af\u09be\u09ac\u09c7') && w.length > 5)
+        return w.slice(0, -4); // যাবে (future)
+    if (w.endsWith('\u0995\u09b0\u09be') && w.length > 4)
+        return w.slice(0, -3); // করা (infinitive)
+    if (w.endsWith('\u0995\u09b0\u09c7') && w.length > 4)
+        return w.slice(0, -3); // করে (present)
+    if (w.endsWith('\u09b9\u09df') && w.length > 3)
+        return w.slice(0, -2); // হয় (is/becomes)
+    if (w.endsWith('\u09b9\u09ac\u09c7') && w.length > 4)
+        return w.slice(0, -3); // হবে (will be)
+    return w;
+}
+/**
+ * Pre-process text BEFORE tokenization to split code identifiers.
+ * Handles camelCase, PascalCase, snake_case, and kebab-case.
+ *
+ * Examples:
+ *   deployServer       → deploy Server  (then lowercased → deploy server)
+ *   HTMLParser         → HTML Parser
+ *   my_api_key         → my api key
+ *   get-user-by-id     → get user by id (hyphens preserved in regex, spaces also fine)
+ *   BackendServiceImpl → Backend Service Impl
+ */
+function preprocessText(text) {
+    return text
+        // camelCase boundary: lowercase/digit → uppercase
+        .replace(/([a-z0-9])([A-Z])/g, '$1 $2')
+        // Consecutive uppercase → uppercase+lowercase boundary: HTMLParser → HTML Parser
+        .replace(/([A-Z]{2,})([A-Z][a-z])/g, '$1 $2')
+        // snake_case: replace underscores with spaces
+        .replace(/_/g, ' ');
+}
 // Maps tokens (English, Japanese katakana, Chinese, Korean, Arabic, Hebrew)
 // to their equivalents in other languages. Used to expand tokens at search time
 // so "deploy" finds documents containing "デプロイ" / "部署" / "배포", and vice versa.
 //
 // Keys must be lowercase. CJK values will be bigram-expanded at use time.
 const CROSS_LINGUAL_MAP = new Map([
+    // ── Tech / Ecosystem synonyms (language-agnostic) ─────────────────────────
+    ['javascript', ['js', 'node', 'nodejs', 'ecmascript']],
+    ['js', ['javascript', 'node', 'nodejs']],
+    ['typescript', ['ts', 'javascript', 'js', 'tsc']],
+    ['ts', ['typescript', 'javascript', 'js']],
+    ['python', ['py', 'pip', 'django', 'flask', 'fastapi']],
+    ['py', ['python', 'pip']],
+    ['golang', ['go', 'gopher', 'goroutine']],
+    ['go', ['golang', 'goroutine']],
+    ['rust', ['cargo', 'rustlang', 'crates']],
+    ['cargo', ['rust', 'rustlang']],
+    ['java', ['jvm', 'maven', 'gradle', 'spring']],
+    ['kotlin', ['jvm', 'android', 'coroutine']],
+    ['dotnet', ['csharp', 'asp', 'nuget', 'aspnet']],
+    ['postgresql', ['postgres', 'pg', 'psql']],
+    ['postgres', ['postgresql', 'pg', 'psql']],
+    ['mysql', ['mariadb', 'sql']],
+    ['mongodb', ['mongo', 'nosql', 'bson', 'mongoose']],
+    ['mongo', ['mongodb', 'nosql', 'bson']],
+    ['elasticsearch', ['elastic', 'opensearch', 'lucene', 'kibana']],
+    ['elastic', ['elasticsearch', 'opensearch']],
+    ['kubernetes', ['k8s', 'kube', 'kubectl', 'helm', 'k3s']],
+    ['k8s', ['kubernetes', 'kube', 'kubectl', 'helm']],
+    ['helm', ['kubernetes', 'k8s', 'chart']],
+    ['terraform', ['tf', 'hcl', 'iac', 'opentofu']],
+    ['ansible', ['playbook', 'automation']],
+    ['grafana', ['dashboard', 'visualization', 'metrics']],
+    ['prometheus', ['metrics', 'alert', 'scrape', 'grafana']],
+    ['nginx', ['proxy', 'webserver', 'ingress']],
+    ['traefik', ['proxy', 'ingress', 'router']],
+    ['rabbitmq', ['amqp', 'queue', 'broker', 'messaging']],
+    ['kafka', ['messaging', 'stream', 'broker']],
+    ['grpc', ['protobuf', 'proto', 'rpc']],
+    ['graphql', ['gql', 'resolver', 'schema']],
+    ['react', ['jsx', 'hooks', 'component', 'redux']],
+    ['nextjs', ['next', 'react', 'ssr', 'vercel']],
+    ['vue', ['vuejs', 'vite', 'nuxt']],
+    ['angular', ['ng', 'rxjs', 'typescript']],
+    ['aws', ['amazon', 'ec2', 's3', 'lambda', 'cloudwatch', 'ecs', 'eks']],
+    ['gcp', ['google cloud', 'gke', 'bigquery']],
+    ['azure', ['microsoft cloud', 'aks', 'devops']],
+    ['s3', ['bucket', 'object storage', 'aws']],
+    ['lambda', ['serverless', 'function', 'faas', 'aws']],
+    ['jwt', ['token', 'bearer', 'oauth', 'auth']],
+    ['oauth', ['oidc', 'auth', 'token', 'sso']],
+    ['webhook', ['callback', 'event', 'trigger']],
+    ['cron', ['schedule', 'job', 'timer']],
+    ['yaml', ['yml', 'config', 'manifest']],
+    ['dotenv', ['env', 'environment', 'envfile']],
+    ['github', ['git', 'actions', 'repo', 'ci']],
+    ['gitlab', ['git', 'pipeline', 'runner']],
+    ['wireguard', ['vpn', 'tunnel', 'wg']],
+    ['prisma', ['orm', 'database', 'schema', 'migration']],
+    ['npm', ['node', 'package', 'registry', 'yarn', 'pnpm']],
+    ['yarn', ['npm', 'package', 'node']],
+    ['pip', ['python', 'package', 'pypi']],
+    ['brew', ['homebrew', 'macos', 'package']],
     // English → all
     ['deploy', ['デプロイ', '部署', '배포', 'deployment', 'deploying', 'نشر', 'פריסה', 'استقرار', 'तैनाती', 'bereitstellen', 'bereitstellung', 'déployer', 'déploiement', 'desplegar', 'despliegue', 'distribuire', 'distribuzione', 'implantar', 'implantação', 'развёртывание', 'развертывание', 'dağıtım', 'dağıtmak']],
     ['deployment', ['デプロイ', '部署', '배포', 'deploy', 'نشر', 'פריסה', 'استقرار', 'तैनाती', 'bereitstellung', 'déploiement', 'despliegue', 'distribuzione', 'implantação', 'развёртывание', 'dağıtım']],
@@ -1392,7 +1533,8 @@ function expandCrossLingual(token) {
  */
 function tokenize(text) {
     const tokens = [];
-    const lower = text.toLowerCase();
+    // Pre-process: split camelCase/PascalCase identifiers and snake_case before lowercasing
+    const lower = preprocessText(text).toLowerCase();
     for (const [seg] of lower.matchAll(SEGMENT_RE)) {
         if (CJK_RE.test(seg)) {
             // CJK segment — extract overlapping character bigrams
@@ -1446,10 +1588,8 @@ function tokenize(text) {
         }
         else if (DEVANAGARI_RE.test(seg)) {
             // Devanagari segment (Hindi / Marathi) — naturally space-separated
-            // Words don't need bigram expansion; stems reduce inflected forms.
             const words = seg.trim().split(/[\s\u200c\u200d]+/); // handle zero-width joiners
             for (const w of words) {
-                // Keep Devanagari chars + digits; discard pure Latin portions (e.g. embedded numbers)
                 const deva = w.replace(/[^\u0900-\u097f0-9]/g, '');
                 if (deva.length <= 1 || STOPWORDS.has(deva))
                     continue;
@@ -1458,6 +1598,18 @@ function tokenize(text) {
                     tokens.push(stemmed);
             }
         }
+        else if (BENGALI_RE.test(seg)) {
+            // Bengali (Bangla) segment — space-separated words, Bengali script
+            const words = seg.trim().split(/[\s\u200c\u200d]+/);
+            for (const w of words) {
+                const bn = w.replace(/[^\u0980-\u09ff0-9]/g, '');
+                if (bn.length <= 1 || STOPWORDS.has(bn))
+                    continue;
+                const stemmed = bengaliLightStem(bn);
+                if (stemmed.length > 1 && !STOPWORDS.has(stemmed))
+                    tokens.push(stemmed);
+            }
+        }
         else if (CYRILLIC_RE.test(seg)) {
             // Cyrillic segment (Russian, Bulgarian, Ukrainian, Serbian …)
             const words = seg.trim().split(/\s+/);
@@ -1471,9 +1623,10 @@ function tokenize(text) {
             }
         }
         else {
-            // Latin / other (includes Turkish with special chars ğ/ş/ı)
+            // Latin / other — includes Turkish (ğ/ş/ı), Polish (ą/ę/ł/ń/ó/ś/ź/ż/ć),
+            // Czech (č/š/ž/ě/ů/ř), Hungarian (ő/ű), Romanian (ș/ț)
             const words = seg
-                .replace(/[^a-záàâãäåæçéèêëíìîïñóòôõöúùûüýÿßœğşıİ0-9:_\-./]/g, ' ')
+                .replace(/[^a-záàâãäåæçéèêëíìîïñóòôõöúùûüýÿßœğşıİąćęłńśźżčšžěůřőűșț0-9:\-./]/g, ' ')
                 .split(/\s+/)
                 .filter(w => w.length > 1 && !STOPWORDS.has(w));
             // Turkish: apply light stemmer to words with Turkish-specific chars
@@ -1527,6 +1680,12 @@ function tokenize(text) {
                 if (stemmed.length > 1 && !STOPWORDS.has(stemmed))
                     expansions.push(stemmed);
             }
+            else if (BENGALI_RE.test(syn)) {
+                // Bengali synonym
+                const stemmed = bengaliLightStem(syn);
+                if (stemmed.length > 1 && !STOPWORDS.has(stemmed))
+                    expansions.push(stemmed);
+            }
             else if (CYRILLIC_RE.test(syn)) {
                 // Cyrillic synonym (Russian)
                 const cyr = syn.replace(/[^\u0400-\u04ff]/g, '');
@@ -1706,7 +1865,9 @@ async function keywordSearch(redis, patterns, query, topK = 10) {
             bigrams.add(`${tokens[j]}|${tokens[j + 1]}`);
         }
         const timestamp = extractTimestamp(content);
-        docs.push({ key: allKeys[i], content, tokens, tokenFreq, bigrams, timestamp });
+        // Key tokens for title-boost: terms appearing in the Redis key get extra weight
+        const keyTokens = new Set(tokenize(allKeys[i]));
+        docs.push({ key: allKeys[i], content, tokens, tokenFreq, bigrams, keyTokens, timestamp });
         totalTokens += tokens.length;
     }
     if (docs.length === 0)
@@ -1812,6 +1973,18 @@ async function keywordSearch(redis, patterns, query, topK = 10) {
             if (sq.length >= 4 && doc.content.toLowerCase().includes(sq.toLowerCase())) {
                 score *= 2.0;
             }
+            // Key/title boost: if query terms appear in the Redis key (=title), 1.5× boost
+            // Keys often encode the primary topic (e.g. "deploy:api:server"), so key hits are high-precision.
+            if (score > 0) {
+                let keyHits = 0;
+                for (const qt of queryTokens) {
+                    if (doc.keyTokens.has(qt))
+                        keyHits++;
+                }
+                if (keyHits > 0) {
+                    score *= 1 + 0.5 * (keyHits / queryTokens.length);
+                }
+            }
             // Recency boost
             score *= recencyBoost(doc.timestamp);
             if (score > 0) {