better observability

Author: arielweinberger

src/index.ts

@@ -2,13 +2,21 @@ import { MCPServer } from "mcp-framework";
 import { config } from "dotenv";
 import path from "path";
 import { initLibrariesWithFeatures } from "./lib/utils/process-libraries.js";
+import { vectorStore } from "./lib/vector-store.js";
 
 config({ path: path.join(process.cwd(), ".env") });
 
 const port = parseInt(process.env.PORT ?? "1234");
 
 await initLibrariesWithFeatures();
 
+async function inspectVectorStore() {
+  const count = await vectorStore.describeIndex({ indexName: "docs" });
+  console.log("Vector store count:", count);
+}
+
+await inspectVectorStore();
+
 const server = new MCPServer({
   transport: {
     type: "http-stream",

src/lib/utils/query-vector-store.ts

@@ -3,47 +3,69 @@ import { getDocsFileContent } from "./content.js";
 import { embed } from "ai";
 import { openai } from "@ai-sdk/openai";
 
-const MIN_SCORE = process.env.MIN_SCORE ? parseFloat(process.env.MIN_SCORE) : 0.25;
+const MIN_SCORE = process.env.MIN_SCORE
+  ? parseFloat(process.env.MIN_SCORE)
+  : 0.25;
 
 if (!process.env.OPENAI_API_KEY) {
   throw new Error("OPENAI_API_KEY is not set");
 }
 
 export async function queryVectorStore(query: string) {
   console.log(`Querying vector store with query: ${query}`);
-  const queryEmbedding = await embed({
-    model: openai.embedding("text-embedding-3-small"),
-    value: query,
-    maxRetries: 3,
-  });
-
-  const results = await vectorStore.query({
-    indexName: "docs" as any,
-    queryVector: queryEmbedding.embedding,
-    topK: 5,
-    minScore: MIN_SCORE,
-  } as any);
+
+  console.log("Creating embeddings...");
+
+  let queryEmbedding: Awaited<ReturnType<typeof embed>>["embedding"] = [];
+  let results: Awaited<ReturnType<typeof vectorStore.query>> = [];
+
+  try {
+    const queryEmbeddingResult = await embed({
+      model: openai.embedding("text-embedding-3-small"),
+      value: query,
+      maxRetries: 3,
+    });
+    queryEmbedding = queryEmbeddingResult.embedding;
+  } catch (error) {
+    console.error("Error creating embeddings:", error);
+    throw error;
+  }
+
+  try {
+    results = await vectorStore.query({
+      indexName: "docs" as any,
+      queryVector: queryEmbedding,
+      topK: 5,
+      minScore: MIN_SCORE,
+    } as any);
+  } catch (error) {
+    console.error("Error querying vector store:", error);
+    throw error;
+  }
 
   console.log(`Retrieved ${results.length} results`);
 
-  const contents = await Promise.all(results.map(async (result) => {
-    const metadata = result.metadata as VectorStoreMetadata;
-    // const fullDocPage = await getDocsFileContent(metadata.webPath);
-    // const { body } = fullDocPage;
-    const fullDocPage = metadata.content;
-
-    return {
-      path: `${metadata.webPath}`,
-      score: parseFloat(result.score.toFixed(3)),
-      title: metadata.title,
-      description: metadata.description?? "",
-      content: fullDocPage,
-    };
-  }));
+  const contents = await Promise.all(
+    results.map(async (result) => {
+      const metadata = result.metadata as VectorStoreMetadata;
+      // const fullDocPage = await getDocsFileContent(metadata.webPath);
+      // const { body } = fullDocPage;
+      const fullDocPage = metadata.content;
+
+      return {
+        path: `${metadata.webPath}`,
+        score: parseFloat(result.score.toFixed(3)),
+        title: metadata.title,
+        description: metadata.description ?? "",
+        content: fullDocPage,
+      };
+    })
+  );
 
   // Remove contents with duplicate paths
-  const uniqueContents = contents.filter((content, index, self) =>
-    index === self.findIndex((t) => t.path === content.path)
+  const uniqueContents = contents.filter(
+    (content, index, self) =>
+      index === self.findIndex((t) => t.path === content.path)
   );
 
   return uniqueContents;