nextlevelbuilder · MelodyBorg · Apr 18, 2026
diff --git a/agents/data-analyst-session.ts b/agents/data-analyst-session.ts
@@ -0,0 +1,147 @@
+/**
+ * RUNTIME — run on every invocation.
+ *
+ * Usage:
+ *   DATA_ANALYST_AGENT_ID=<id> \
+ *   DATA_ANALYST_ENVIRONMENT_ID=<id> \
+ *   npx ts-node agents/data-analyst-session.ts "What are the top 5 events by volume?"
+ *
+ * Required env vars (from data-analyst-setup.ts output):
+ *   ANTHROPIC_API_KEY
+ *   DATA_ANALYST_AGENT_ID
+ *   DATA_ANALYST_ENVIRONMENT_ID
+ *
+ * Optional:
+ *   DATA_ANALYST_AGENT_VERSION   — pin to a specific agent version
+ *   DATA_ANALYST_VAULT_ID        — vault containing Amplitude MCP credentials
+ */
+
+import Anthropic from "@anthropic-ai/sdk";
+
+const AGENT_ID = process.env.DATA_ANALYST_AGENT_ID;
+const AGENT_VERSION = process.env.DATA_ANALYST_AGENT_VERSION
+  ? Number(process.env.DATA_ANALYST_AGENT_VERSION)
+  : undefined;
+const ENVIRONMENT_ID = process.env.DATA_ANALYST_ENVIRONMENT_ID;
+const VAULT_ID = process.env.DATA_ANALYST_VAULT_ID;
+
+if (!AGENT_ID || !ENVIRONMENT_ID) {
+  console.error(
+    "Missing required env vars: DATA_ANALYST_AGENT_ID, DATA_ANALYST_ENVIRONMENT_ID\n" +
+      "Run agents/data-analyst-setup.ts first.",
+  );
+  process.exit(1);
+}
+
+const userMessage = process.argv.slice(2).join(" ") || "Hello! What can you help me analyse?";
+
+async function runSession() {
+  const client = new Anthropic();
+
+  // 1. Create session referencing the pre-created agent
+  console.log(`Starting session with agent ${AGENT_ID}...`);
+  const session = await client.beta.sessions.create({
+    agent: AGENT_VERSION
+      ? { type: "agent", id: AGENT_ID, version: AGENT_VERSION }
+      : AGENT_ID,
+    environment_id: ENVIRONMENT_ID,
+    title: `Data analysis — ${new Date().toISOString()}`,
+    ...(VAULT_ID ? { vault_ids: [VAULT_ID] } : {}),
+  });
+  console.log(`Session ${session.id} created (status: ${session.status})\n`);
+
+  // 2. Open stream BEFORE sending the message (stream-first ordering)
+  const stream = await client.beta.sessions.stream(session.id);
+
+  await client.beta.sessions.events.send(session.id, {
+    events: [
+      {
+        type: "user.message",
+        content: [{ type: "text", text: userMessage }],
+      },
+    ],
+  });
+
+  // 3. Process the event stream
+  const customToolCalls: Array<{ id: string; tool_name: string; input: unknown }> = [];
+
+  for await (const event of stream) {
+    switch (event.type) {
+      case "agent.message":
+        for (const block of event.content) {
+          if (block.type === "text") {
+            process.stdout.write(block.text);
+          }
+        }
+        break;
+
+      case "agent.custom_tool_use":
+        // Custom tool calls — send results back to continue the session
+        customToolCalls.push({ id: event.id, tool_name: event.tool_name, input: event.input });
+        break;
+
+      case "session.status_idle":
+        if (event.stop_reason?.type === "requires_action") {
+          // Session is waiting for custom tool results — handle them
+          if (customToolCalls.length > 0) {
+            const results = customToolCalls.splice(0).map((call) => ({
+              type: "user.custom_tool_result" as const,
+              custom_tool_use_id: call.id,
+              content: [
+                {
+                  type: "text" as const,
+                  text: `Tool "${call.tool_name}" is not implemented in this runner. Input was: ${JSON.stringify(call.input)}`,
+                },
+              ],
+            }));
+            await client.beta.sessions.events.send(session.id, { events: results });
+          }
+        } else {
+          // Terminal idle — end_turn or retries_exhausted
+          console.log("\n\n[Session complete]");
+        }
+        break;
+
+      case "session.status_terminated":
+        console.log("\n[Session terminated]");
+        break;
+
+      case "session.error":
+        console.error("\n[Session error]", event);
+        break;
+
+      case "span.model_request_end":
+        // Log token usage for cost tracking
+        if (event.model_usage) {
+          const u = event.model_usage;
+          console.error(
+            `\n[usage] input=${u.input_tokens} output=${u.output_tokens}` +
+              ` cache_read=${u.cache_read_input_tokens} cache_write=${u.cache_creation_input_tokens}`,
+          );
+        }
+        break;
+    }
+  }
+
+  // 4. List any files the agent wrote to /mnt/session/outputs/
+  try {
+    const files = await client.beta.files.list({
+      scope_id: session.id,
+      betas: ["managed-agents-2026-04-01"],
+    } as Parameters<typeof client.beta.files.list>[0]);
+
+    if (files.data.length > 0) {
+      console.log("\n[Output files]");
+      for (const f of files.data) {
+        console.log(`  ${f.filename} (${f.size_bytes} bytes) — file_id: ${f.id}`);
+      }
+    }
+  } catch {
+    // scope_id filtering may not be available in all SDK versions
+  }
+}
+
+runSession().catch((err) => {
+  console.error(err);
+  process.exit(1);
+});
diff --git a/agents/data-analyst-setup.ts b/agents/data-analyst-setup.ts
@@ -0,0 +1,111 @@
+/**
+ * ONE-TIME SETUP — run once, save the printed IDs to your .env or config.
+ *
+ * Usage:
+ *   npx ts-node agents/data-analyst-setup.ts
+ *   # or: bun agents/data-analyst-setup.ts
+ *
+ * Required env vars:
+ *   ANTHROPIC_API_KEY
+ *
+ * Optional (for Amplitude MCP auth):
+ *   AMPLITUDE_ACCESS_TOKEN   — OAuth access token for https://mcp.amplitude.com/mcp
+ *   AMPLITUDE_REFRESH_TOKEN  — OAuth refresh token (enables auto-refresh)
+ */
+
+import Anthropic from "@anthropic-ai/sdk";
+
+const SYSTEM_PROMPT = `You analyze data. Given a dataset (file path, URL, or query) and a question:
+
+1. Load the data and print its shape, column names, dtypes, and a small sample. Always look before you compute.
+2. Clean obvious issues — nulls, duplicates, type mismatches — and note what you changed.
+3. Answer the question with code. Prefer pandas/polars for tabular work, matplotlib/plotly for charts. Show intermediate results so your reasoning is checkable.
+4. For product-analytics questions, query Amplitude directly — event funnels, retention cohorts, property breakdowns — and link the chart.
+5. Save any charts or derived tables to /mnt/session/outputs/ and summarize findings in plain language, including caveats (sample size, missing data, correlation-vs-causation).
+
+Default to simple, readable analysis over clever one-liners. A clear bar chart usually beats a dense heatmap.`;
+
+const AMPLITUDE_MCP_URL = "https://mcp.amplitude.com/mcp";
+
+async function setup() {
+  const client = new Anthropic();
+
+  // 1. Create environment
+  console.log("Creating environment...");
+  const environment = await client.beta.environments.create({
+    name: "data-analyst-env",
+    config: {
+      type: "cloud",
+      networking: { type: "unrestricted" },
+    },
+  });
+  console.log(`✓ Environment created: ${environment.id}`);
+
+  // 2. Optionally create a vault for Amplitude MCP credentials
+  let vaultId: string | undefined;
+  const accessToken = process.env.AMPLITUDE_ACCESS_TOKEN;
+
+  if (accessToken) {
+    console.log("Creating vault for Amplitude credentials...");
+    const vault = await client.beta.vaults.create({ name: "data-analyst-vault" });
+
+    await client.beta.vaults.credentials.create(vault.id, {
+      display_name: "Amplitude MCP",
+      auth: {
+        type: "mcp_oauth",
+        mcp_server_url: AMPLITUDE_MCP_URL,
+        access_token: accessToken,
+        ...(process.env.AMPLITUDE_REFRESH_TOKEN
+          ? {
+              refresh: {
+                refresh_token: process.env.AMPLITUDE_REFRESH_TOKEN,
+                // Update these to match your OAuth app registration
+                client_id: process.env.AMPLITUDE_CLIENT_ID ?? "",
+                token_endpoint: "https://api.amplitude.com/oauth2/token",
+                token_endpoint_auth: { type: "none" },
+              },
+            }
+          : {}),
+      },
+    });
+
+    vaultId = vault.id;
+    console.log(`✓ Vault created: ${vault.id}`);
+  } else {
+    console.log(
+      "⚠ AMPLITUDE_ACCESS_TOKEN not set — skipping vault creation.\n" +
+        "  Set it and re-run, or attach credentials manually before starting sessions.",
+    );
+  }
+
+  // 3. Create the agent
+  console.log("Creating Data analyst agent...");
+  const agent = await client.beta.agents.create({
+    name: "Data analyst",
+    model: "claude-sonnet-4-6",
+    system: SYSTEM_PROMPT,
+    mcp_servers: [
+      { type: "url", name: "amplitude", url: AMPLITUDE_MCP_URL },
+    ],
+    tools: [
+      { type: "agent_toolset_20260401", default_config: { enabled: true } },
+      { type: "mcp_toolset", mcp_server_name: "amplitude" },
+    ],
+  });
+  console.log(`✓ Agent created: ${agent.id}  (version: ${agent.version})`);
+
+  // 4. Print env vars to persist
+  console.log("\n─── Add these to your .env ───");
+  console.log(`DATA_ANALYST_AGENT_ID=${agent.id}`);
+  console.log(`DATA_ANALYST_AGENT_VERSION=${agent.version}`);
+  console.log(`DATA_ANALYST_ENVIRONMENT_ID=${environment.id}`);
+  if (vaultId) {
+    console.log(`DATA_ANALYST_VAULT_ID=${vaultId}`);
+  }
+  console.log("──────────────────────────────");
+}
+
+setup().catch((err) => {
+  console.error(err);
+  process.exit(1);
+});