memoryai-mcp 2.4.1 → 2.4.3

package/README.md

@@ -344,7 +344,7 @@ DNA memories (preference/decision/identity) **never decay, never get deleted, ne
 | Free | Basic store/recall, 100 memories | Free |
 | Pro | Full brain (reasoning, consolidation, personality) | Paid |
 | ProMax | Multi-agent mesh, advanced features | Paid |
-| God | Everything + deep graph traversal | Internal |
+| Enterprise | Everything + deep graph traversal + on-prem | Custom |
 
 Get started free: https://memoryai.dev
 

package/dist/index.js

@@ -11,7 +11,7 @@ import { z } from "zod";
 const API_URL = process.env.MEMORYAI_ENDPOINT || process.env.HM_ENDPOINT || "http://localhost:8420";
 const API_KEY = process.env.MEMORYAI_API_KEY || process.env.HM_API_KEY || "";
 const REQUEST_TIMEOUT_MS = 30_000; // P2 #6: 30s default timeout for API requests
-const MCP_VERSION = "2.4.1";
+const MCP_VERSION = "2.4.3";
 // Context Guard — per-IDE settings via env vars.
 // HM_COMPACT_AT and HM_CRITICAL_AT are now ABSOLUTE token counts (e.g. "100000",
 // "150000"). The legacy meaning ("30" = 30%) is detected automatically: any
@@ -24,6 +24,13 @@ const MCP_VERSION = "2.4.1";
 const CG_CONTEXT_CAP = parseInt(process.env.MEMORYAI_CONTEXT_CAP || process.env.HM_CONTEXT_CAP || "0", 10);
 const CG_COMPACT_RAW = parseInt(process.env.MEMORYAI_COMPACT_AT || process.env.HM_COMPACT_AT || "0", 10);
 const CG_CRITICAL_RAW = parseInt(process.env.MEMORYAI_CRITICAL_AT || process.env.HM_CRITICAL_AT || "0", 10);
+// Model hint for server-side window auto-detection. When set, the server
+// resolves the context window from the model name and picks the adaptive
+// trigger percentage (<=200K → 95%, >200K → 30%). Falls back to the model
+// the caller passes per-request, then to the 200K default.
+//   e.g. MEMORYAI_MODEL=claude-opus-4-6[1m]  → 1M window → 30% trigger
+//        MEMORYAI_MODEL=claude-sonnet-4-6     → 200K window → 95% trigger
+const CG_MODEL = (process.env.MEMORYAI_MODEL || process.env.HM_MODEL || "").trim() || null;
 // Heuristic: small numbers are legacy percentages; large numbers are absolute tokens.
 // Threshold "<= 100" is generous enough to catch any sensible % (max 95%) and
 // well below any sensible absolute count (min would be ~10K tokens).
@@ -152,7 +159,7 @@ function err(e) {
     return { content, isError: true };
 }
 // --- MCP Server ---
-const server = new McpServer({ name: "memoryai", version: "2.4.1" }, {
+const server = new McpServer({ name: "memoryai", version: "2.4.3" }, {
     capabilities: { tools: {} },
     instructions: "MemoryAI persistent memory. Call memory_bootstrap on session start. After decisions/preferences, call memory_store. Context compaction is automatic via piggybacking — follow any [Context Guard] directives in tool responses.",
 });
@@ -1013,12 +1020,13 @@ server.tool("context_guard_check", "[CORE] Check context pressure — returns re
     model: z.string().optional().describe("Model name for auto-detecting context window size (e.g. claude-sonnet-4-6)"),
 }, async (args) => {
     try {
-        // Use env var HM_CONTEXT_CAP as default if max_tokens not provided
+        // Use env var HM_CONTEXT_CAP as default if max_tokens not provided.
+        // 0 → server auto-detects the window from the model name.
         const maxTokens = args.max_tokens || CG_CONTEXT_CAP || 0;
         const payload = {
             estimated_tokens: args.estimated_tokens,
             max_tokens: maxTokens,
-            model: args.model || null,
+            model: args.model || CG_MODEL,
         };
         // Per-IDE threshold overrides. Absolute (CG_*_AT_TOKENS) is preferred —
         // server treats it as the authoritative trigger. Decimal % is the
@@ -1097,7 +1105,7 @@ server.tool("ide_turn_check", "[CORE] Server-authoritative context check for IDE
             max_tokens: args.max_tokens ?? CG_CONTEXT_CAP ?? 200_000,
             avg_tokens_per_turn: args.avg_tokens_per_turn ?? 8000,
             skip_below_turns: args.skip_below_turns ?? 10,
-            model: args.model ?? null,
+            model: args.model ?? CG_MODEL,
         };
         // Per-IDE threshold overrides. Absolute first (preferred), % fallback.
         if (CG_COMPACT_AT_TOKENS > 0)
@@ -1210,7 +1218,7 @@ server.tool("bot_guard_check", "Advanced: Bot context guard — checks context p
         const payload = {
             estimated_tokens: args.estimated_tokens,
             max_tokens: args.max_tokens || CG_CONTEXT_CAP || 200000,
-            model: args.model || null,
+            model: args.model || CG_MODEL,
         };
         if (args.compress_threshold)
             payload.compress_threshold = args.compress_threshold;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "memoryai-mcp",
-  "version": "2.4.1",
+  "version": "2.4.3",
   "description": "MCP server for MemoryAI v2.3 — One brain. Every AI you use. Forever. Persistent memory and context guard tools for IDEs and bots.",
   "homepage": "https://memoryai.dev",
   "repository": {