- Added a PreWarm Model feature to the MCP server. I am using this in a rather large NX monorepo, and tools like search_codebase were timing out, and the indexing was timing out. What this PreWarm feature does is, as the tool initializes the MCP server, we index the codebase so that it's ready when the agent needs to use the codebase context mcp server for it's tools.

Author: aolin480

src/index.ts

@@ -1133,6 +1133,20 @@ async function main() {
     logStartup('[DEBUG] Index found. Ready.');
     indexState.status = 'ready';
     indexState.lastIndexed = new Date();
+    
+    // Pre-warm the embedding model in background to avoid timeout on first search
+    // This loads the ML model (can take 30-90s) before any search request comes in
+    if (process.env.CODEBASE_CONTEXT_PREWARM_MODEL !== 'false') {
+      logStartup('[DEBUG] Pre-warming embedding model...');
+      import('./embeddings/index.js')
+        .then(({ getEmbeddingProvider }) => getEmbeddingProvider())
+        .then(() => {
+          logStartup('[DEBUG] Embedding model ready');
+        })
+        .catch((err) => {
+          console.warn('[WARN] Failed to pre-warm embedding model:', err.message);
+        });
+    }
   }
 
   const transport = new StdioServerTransport();