@deepsweet/mdn 0.1.2 → 0.2.0

package/README.md

@@ -22,7 +22,7 @@ See [dataset repo](https://huggingface.co/datasets/deepsweet/mdn) on HuggigFace
 ### 1. Download dataset and embedding model
 
 ```sh
-npx -y @deepsweet/mdn download
+npx -y @deepsweet/mdn@latest download
 ```
 
 Both [dataset](https://huggingface.co/datasets/deepsweet/mdn) (\~260 MB) and the [embedding model GGUF file](https://huggingface.co/deepsweet/bge-m3-GGUF-Q4_K_M) (\~438 MB) will be downloaded directly from HugginFace and stored in its default cache location (typically `~/.cache/huggingface/`), just like the `hf download` command does.
@@ -35,7 +35,8 @@ Both [dataset](https://huggingface.co/datasets/deepsweet/mdn) (\~260 MB) and the
     "mdn": {
       "command": "npx",
       "args": [
-        "@deepsweet/mdn",
+        "-y",
+        "@deepsweet/mdn@latest",
         "server"
       ],
       "env": {}
@@ -44,6 +45,9 @@ Both [dataset](https://huggingface.co/datasets/deepsweet/mdn) (\~260 MB) and the
 }
 ```
 
+> [!TIP]
+> Remove `@latest` for a full offline experience, but keep in mind that this will cache a fixed version without auto-updating.
+
 The `stdio` server will spawn [llama.cpp](https://github.com/ggml-org/llama.cpp) under the hood, load the embedding model (~655 MB RAM/VRAM), and query the dataset – all on demand.
 
 ## Settings

package/dist/index.js

@@ -105,6 +105,9 @@ var getLlamaContext = async (modelPath) => {
     batchSize: MAX_TOKENS,
     threads: 0
   });
+  context.onDispose.createOnceListener(() => {
+    model.dispose().then(() => llama.dispose()).catch(console.error);
+  });
   return context;
 };
 
@@ -131,7 +134,7 @@ var createReranker = async () => {
 
 // package.json
 var name = "@deepsweet/mdn";
-var version = "0.1.1";
+var version = "0.2.0";
 
 // src/server.ts
 var startMcpServer = async () => {
@@ -149,6 +152,9 @@ var startMcpServer = async () => {
     description: "Reference documentation for Web API, JavaScript, HTML, CSS, SVG and HTTP",
     inputSchema: z2.object({
       query: z2.string().describe(env.MDN_QUERY_DESCRIPTION)
+    }),
+    outputSchema: z2.object({
+      results: z2.array(z2.string())
     })
   }, async ({ query }) => {
     llamaContext ??= await getLlamaContext(modelPath);
@@ -167,10 +173,29 @@ var startMcpServer = async () => {
       content: results.map((result) => ({
         type: "text",
         text: result.text
-      }))
+      })),
+      structuredContent: {
+        results: results.map((result) => result.text)
+      }
     };
   });
   const transport = new StdioServerTransport;
+  const dispose = async () => {
+    if (llamaTimeout !== null) {
+      clearTimeout(llamaTimeout);
+    }
+    if (llamaContext !== null) {
+      await llamaContext.dispose();
+    }
+    db.close();
+    await server.close();
+  };
+  transport.onclose = () => {
+    dispose().catch(console.error);
+  };
+  transport.onerror = (err) => {
+    console.log(err);
+  };
   await server.connect(transport);
 };
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@deepsweet/mdn",
-  "version": "0.1.2",
+  "version": "0.2.0",
   "publishConfig": {
     "access": "public"
   },