@cachly-dev/mcp-server 0.2.0 → 0.5.1

package/README.md

@@ -50,7 +50,7 @@ Once connected, just talk to your AI assistant:
 | **`recall_best_solution`** | Retrieve the best known solution for a topic (increments recall count). |
 | **`remember_context`** | Cache any analysis or architecture finding for future sessions. |
 | **`recall_context`** | Retrieve cached context by exact key (supports glob: `"file:*"`). |
-| **`smart_recall`** | Semantic search across all cached context by meaning/keywords. |
+| **`smart_recall`** | BM25+ keyword search across all cached context. Works **without any embedding provider**. Optionally adds semantic vector search if an API key is configured. |
 | **`list_remembered`** | List all cached context entries. |
 | **`forget_context`** | Delete stale context. |
 
@@ -225,23 +225,7 @@ AI:  [calls smart_recall("auth")]
      (No file reading needed — instant answer!)
 ```
 
-**Setup in 2 minutes:**
-
-1. Add to `.github/copilot-instructions.md` (or `CLAUDE.md` for Claude Code):
-```markdown
-## Cachly AI Brain
-
-At the START of every session:
-session_start(instance_id = "<id>", focus = "<what you plan to work on>")
-
-At the END of every session:
-session_end(instance_id = "<id>", summary = "...", files_changed = [...])
-
-After any bug fix: learn_from_attempts(instance_id = "<id>", topic = "category:keyword",
-  outcome = "success", what_worked = "...", severity = "major")
-```
-
-2. From now on: your AI calls `session_start` first and never re-reads what it already knows.
+The `setup` wizard (see top of this README) writes the session instructions automatically into your `CLAUDE.md` / `.github/copilot-instructions.md` — **no manual editing needed.** Just run `npx @cachly-dev/mcp-server setup` once and your AI assistant calls `session_start` at the start of every session automatically.
 
 ### 1. AI-Assisted Development — "Which caches am I using?"
 

package/dist/index.d.ts

@@ -1,2 +1,31 @@
 #!/usr/bin/env node
-export {};
+/**
+ * Stopwords — filtered out during tokenization.
+ * Covers: English, German, French, Spanish, Italian, Portuguese.
+ * Keeps the index small and scores meaningful.
+ */
+declare const STOPWORDS: Set<string>;
+/** Tokenize text into meaningful keywords. Handles unicode (accented chars). */
+declare function tokenize(text: string): string[];
+/**
+ * Split a multi-topic query into sub-queries.
+ * Detects numbered lists, semicolons, "and also", line breaks, etc.
+ *
+ * Example: "deploy API fix routing and also check auth" →
+ *   ["deploy API fix routing", "check auth"]
+ *
+ * Example: "1. deploy 2. fix routing 3. auth" →
+ *   ["deploy", "fix routing", "auth"]
+ */
+declare function splitMultiQuery(query: string): string[];
+/**
+ * Levenshtein distance — edit distance between two strings.
+ * Used for typo-tolerant fuzzy matching (distance ≤ 2 = match).
+ * O(n*m) but strings are short (tokens), so this is fast.
+ */
+declare function levenshtein(a: string, b: string): number;
+/** Extract a timestamp from JSON content (looks for "ts" or "created" fields). */
+declare function extractTimestamp(content: string): number | undefined;
+/** Recency multiplier: 1.0 for now, 0.5 after half-life days, exponential decay. */
+declare function recencyBoost(timestampMs: number | undefined): number;
+export { tokenize, splitMultiQuery, levenshtein, recencyBoost, extractTimestamp, STOPWORDS };