Claude Sample Agent - Node.js

Browse code

This sample demonstrates how to build an agent using Claude in Node.js with the Microsoft Agent 365 SDK. It covers:

  • Observability: Auto-instrumentation via @microsoft/opentelemetry distro with explicit InferenceScope for LLM call tracing
  • Notifications: Services and models for managing user notifications
  • Tools: Model Context Protocol tools for building advanced agent solutions
  • Hosting Patterns: Hosting with Microsoft 365 Agents SDK

This sample uses the Microsoft Agent 365 SDK for Node.js.

For comprehensive documentation and guidance on building agents with the Microsoft Agent 365 SDK, including how to add tooling, observability, and notifications, visit the Microsoft Agent 365 Developer Documentation.

Prerequisites

To run the template in your local dev machine, you will need:

  • Microsoft Agent 365 SDK
  • Claude Agent SDK (@anthropic-ai/claude-agent-sdk)
  • A365 CLI: Required for agent deployment and management.

Configuration

Observability export

src/index.ts imports the observability bootstrap first. With useS2SEndpoint: true and the app-only tokenResolver, OBS posts to /observabilityService/tenants/{tenant}/otlp/agents/{agent}/traces?api-version=1. The resolver uses AGENT365_OBS_TENANT_ID, AGENT365_OBS_AGENT_ID, AGENT365_OBS_BLUEPRINT_CLIENT_ID, and AGENT365_OBS_BLUEPRINT_CLIENT_SECRET; AGENT365_OBS_AGENT_ID must be the actual agent instance client ID. Business MCP, Graph, and OBO calls remain separate from OBS authentication.

Single-instance limitation: this resolver exports for one statically configured agent instance and tenant (AGENT365_OBS_*). It needs its own copy of the blueprint secret and has no managed-identity option. Multi-instance or multi-tenant deployments should reuse the hosting connection per agent/tenant instead of sharing this static provider.

Sovereign clouds are not supported by this sample provider because the authority is hard-coded to login.microsoftonline.com. For AI Teammates, complete the Agent365.Observability.OtelWrite application-role step printed by a365 setup all --aiteammate; AI Teammate S2S without it has not been validated.

Working with User Identity

On every incoming message, the A365 platform populates activity.from with basic user information — always available with no API calls or token acquisition:

Field Description
activity.from.id Channel-specific user ID (e.g., 29:1AbcXyz... in Teams)
activity.from.name Display name as known to the channel
activity.from.aadObjectId Azure AD Object ID — use this to call Microsoft Graph

The sample logs these fields at the start of every message turn and injects the display name into the LLM system instructions for personalized responses.

Running the Agent in Microsoft 365 Agents Playground

  1. First, select the Microsoft 365 Agents Toolkit icon on the left in the VS Code toolbar.
  2. In file env/.env.playground.user, fill in your Anthropic API key SECRET_ANTHROPIC_API_KEY=<your-key>.
  3. In file env/.env.playground, fill in your custom app registration client id CLIENT_APP_ID.
  4. Press F5 to start debugging which launches your agent in Microsoft 365 Agents Playground using a web browser. Select Debug in Microsoft 365 Agents Playground.
  5. You can send any message to get a response from the agent.

Congratulations! You are running an agent that can now interact with users in Microsoft 365 Agents Playground.

Running Locally with Teams

Use this path when testing against real Teams traffic via a dev tunnel instead of Agents Playground.

1. Create your .env file

Copy .env.template to .env in the sample-agent directory and fill in the required values:

cp .env.template .env

Required values:

Variable Description
ANTHROPIC_API_KEY Your Anthropic API key — get one at console.anthropic.com
NODE_ENV Set to production so JWT validation is enabled (Teams always sends auth tokens)
OTEL_SERVICE_NAME Service name shown in traces (e.g. Claude Sample Agent)
connections__service_connection__settings__clientId Blueprint App ID from a365.generated.config.json (agentBlueprintId)
connections__service_connection__settings__clientSecret Blueprint client secret value (not the secret ID)
connections__service_connection__settings__tenantId Your Azure AD tenant ID

Note: NODE_ENV=development (the template default) skips JWT validation entirely, which works for Agents Playground (no auth header) but fails for Teams (always sends a JWT). Set NODE_ENV=production when testing with real Teams traffic.

2. Start the agent

npm run dev

The agent listens on http://127.0.0.1:3978. You should see:

Server listening on 127.0.0.1:3978 for appId <blueprint-id> debug agents:*

3. Start a dev tunnel

devtunnel host <your-tunnel-name> --allow-anonymous

If port 3978 is not yet mapped to the tunnel:

devtunnel port create <your-tunnel-name> -p 3978 --protocol https

4. Test in Teams

Send the agent a message. The agent only handles new messages (message activity type) — do not edit a sent message, as Teams sends a messageUpdate activity for edits which has no handler.

Handling Agent Install and Uninstall

When a user installs (hires) or uninstalls (removes) the agent, the A365 platform sends an InstallationUpdate activity — also referred to as the agentInstanceCreated event. The sample handles this in handleInstallationUpdateActivity (agent.ts):

Action Description
add Agent was installed — send a welcome message
remove Agent was uninstalled — send a farewell message 
if (context.activity.action === 'add') {
  await context.sendActivity('Thank you for hiring me! Looking forward to assisting you in your professional journey!');
} else if (context.activity.action === 'remove') {
  await context.sendActivity('Thank you for your time, I enjoyed working with you.');
}

To test with Agents Playground, use Mock an Activity → Install application to send a simulated installationUpdate activity.

Sending Multiple Messages in Teams

Agent365 agents can send multiple discrete messages in response to a single user prompt in Teams. This is achieved by calling sendActivity multiple times within a single turn.

Important: Streaming responses are not supported for agentic identities in Teams. The SDK detects agentic identity and buffers the stream into a single message. Use sendActivity directly to send immediate, discrete messages to the user.

The sample demonstrates this in handleAgentMessageActivity (agent.ts):

// Message 1: immediate ack — reaches the user right away
await turnContext.sendActivity('Got it — working on it…');

// ... LLM processing ...

// Message 2: the LLM response
await turnContext.sendActivity(response);

Each sendActivity call produces a separate Teams message. You can call it as many times as needed to send progress updates, partial results, or a final answer.

Typing Indicators

The agent sends typing indicators in a loop every ~4 seconds to keep the ... animation alive while the LLM processes the request:

let typingInterval: ReturnType<typeof setInterval> | undefined;
const startTypingLoop = () => {
  typingInterval = setInterval(async () => {
    await turnContext.sendActivity({ type: 'typing' } as Activity);
  }, 4000);
};
const stopTypingLoop = () => { clearInterval(typingInterval); };

startTypingLoop();
try {
  // ... LLM processing ...
} finally {
  stopTypingLoop();
}

Note: Typing indicators are only visible in 1:1 chats and small group chats — not in channels.

Observability

This sample uses the @microsoft/opentelemetry distro, initialised in src/otel.ts:

import { useMicrosoftOpenTelemetry, shutdownMicrosoftOpenTelemetry } from '@microsoft/opentelemetry';
import { createObservabilityTokenResolver } from './observability-token-service';
useMicrosoftOpenTelemetry({
  a365: { enabled: true, useS2SEndpoint: true, tokenResolver: createObservabilityTokenResolver() },
});

This file is imported first in src/index.ts so instrumentation patches are applied before any HTTP modules load.

Auto-instrumented spans

Span What it captures
POST /api/messages Inbound request — method, path, status code
POST login.microsoftonline.com MSAL token acquisition
POST smba.trafficmanager.net Outbound Teams messages

Why InferenceScope is explicit

The @anthropic-ai/claude-agent-sdk executes inference by spawning the Claude CLI as a child process. The actual HTTPS call to api.anthropic.com happens inside that subprocess — it is invisible to the parent process's HTTP auto-instrumentation. InferenceScope in src/client.ts manually brackets the query() call to produce a span with gen_ai.* attributes (model, provider, agent ID) that would otherwise be missing entirely.

Service Name

Set OTEL_SERVICE_NAME in .env to give your service a meaningful name in traces:

OTEL_SERVICE_NAME=Claude Sample Agent

Running the Agent

To set up and test this agent, refer to the Configure Agent Testing guide for complete instructions.

Troubleshooting

Audience mismatch / 401 Unauthorized from Teams

agents:jwt-middleware:error Audience mismatch <blueprint-id>

Cause: NODE_ENV=development causes the agent to skip loading auth configuration, so JWT audience matching fails when Teams sends a token.

Fix: Set NODE_ENV=production in .env and ensure connections__service_connection__settings__clientId matches your Blueprint App ID.

Error: Claude Code process exited with code 1

Cause: The @anthropic-ai/claude-agent-sdk spawns the claude CLI as a subprocess. If you're developing inside VS Code with the Claude Code extension, the CLAUDECODE environment variable is set in the parent process and inherited by the subprocess, which rejects nested sessions.

Fix: Already handled in src/client.ts — the CLAUDECODE variable is deleted from the subprocess environment before spawning. If you see this error, ensure you are on the latest version of this sample.

Port 3978 already in use

Error: listen EADDRINUSE: address already in use 127.0.0.1:3978

Cause: A previous agent process crashed but did not release the port.

Fix: Kill the occupying process and restart:

# PowerShell
$proc = Get-NetTCPConnection -LocalPort 3978 | Select-Object -ExpandProperty OwningProcess
Stop-Process -Id $proc -Force

# bash / macOS / Linux
lsof -ti:3978 | xargs kill -9

Agent does not respond to an edited message

Cause: Editing a message in Teams sends a messageUpdate activity (eventType: editMessage). The agent only registers a handler for message activities.

Fix: Send a new message instead of editing an existing one. Support for messageUpdate can be added by registering a handler for ActivityTypes.MessageUpdate if needed.

invalid_client when acquiring tokens

AADSTS7000216: 'client_assertion', 'client_secret' or 'request' is required

Cause: connections__service_connection__settings__clientSecret is empty in .env.

Fix: Set the Blueprint client secret value (not the secret ID) in .env.

Support

For issues, questions, or feedback:

Contributing

This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit https://cla.opensource.microsoft.com.

When you submit a pull request, a CLA bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., status check, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repos using our CLA.

This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact opencode@microsoft.com with any additional questions or comments.

Additional Resources

Trademarks

Microsoft, Windows, Microsoft Azure and/or other Microsoft products and services referenced in the documentation may be either trademarks or registered trademarks of Microsoft in the United States and/or other countries. The licenses for this project do not grant you rights to use any Microsoft names, logos, or trademarks. Microsoft's general trademark guidelines can be found at http://go.microsoft.com/fwlink/?LinkID=254653.

License

Copyright (c) Microsoft Corporation. All rights reserved.

Licensed under the MIT License - see the LICENSE file for details.