CrewAI Agent Sample - Python

Browse code

This sample demonstrates how to build a multi-agent system using CrewAI while integrating with the Microsoft Agent 365 SDK. It mirrors the structure and hosting patterns of the AgentFramework/OpenAI Agent 365 samples, while preserving native CrewAI logic in src/crew_agent/.

Demonstrates

  • MCP Tooling: Full Model Context Protocol (MCP) integration with Microsoft 365 services (Mail, Calendar, Copilot)
  • Observability: Complete tracing with InvokeAgentScope, InferenceScope, and ExecuteToolScope for all agent operations
  • Notifications: Email and Teams @mention notification support
  • Multi-Agent Orchestration: Sequential agent workflow with Weather Checker and Driving Safety Advisor agents
  • Azure OpenAI Integration: Native support for Azure OpenAI deployments via CrewAI's azure-ai-inference extra and LiteLLM routing
  • Hosting Patterns: Hosting with Microsoft 365 Agents SDK

This sample uses the Microsoft Agent 365 SDK for Python.

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

  • Python 3.11–3.13
  • UV (recommended for dependency management)
  • Azure OpenAI API credentials OR OpenAI API key
  • Tavily API key for weather search functionality
  • Microsoft 365 Agents Playground (optional, for testing)
  • Bearer token from a365 develop get-token -o raw (for MCP server authentication)

Configuration

Step 1: Create Environment File

Copy .env.template to .env:

cp .env.template .env

Step 2: Configure Required Variables

For Azure OpenAI (recommended):

AZURE_API_KEY=your-azure-openai-key
AZURE_API_BASE=https://your-resource.openai.azure.com/
AZURE_API_VERSION=2025-01-01-preview
OPENAI_API_KEY=your-azure-openai-key  # CrewAI requires this
OPENAI_MODEL_NAME=azure/gpt-4.1       # Use azure/ prefix for LiteLLM

For OpenAI:

OPENAI_API_KEY=sk-your-openai-key
OPENAI_MODEL_NAME=gpt-4o-mini

For MCP Tools (Mail, Calendar, Copilot):

BEARER_TOKEN=<run: a365 develop get-token -o raw>
USE_AGENTIC_AUTH=false
AGENTIC_APP_ID=crewai-agent

For Weather Tool:

TAVILY_API_KEY=tvly-your-tavily-key

Observability S2S export

A365 export is disabled by default. To send traces to Agent 365, set the dedicated OBS credentials and enable the exporter:

ENABLE_A365_OBSERVABILITY_EXPORTER=true
AGENT365_OBS_TENANT_ID=<<YOUR_TENANT_ID>>
AGENT365_OBS_AGENT_ID=<<YOUR_AGENT_INSTANCE_CLIENT_ID>>
AGENT365_OBS_BLUEPRINT_CLIENT_ID=<<YOUR_BLUEPRINT_CLIENT_ID>>
AGENT365_OBS_BLUEPRINT_CLIENT_SECRET=<<YOUR_BLUEPRINT_CLIENT_SECRET>>

AGENT365_OBS_AGENT_ID is the actual runtime agent instance client ID, not the blueprint ID, service-principal object ID, or agent-user ID. ENABLE_A365_OBSERVABILITY_EXPORTER=false disables A365 HTTP export; samples using the Microsoft OpenTelemetry distro can still enrich spans for other exporters.

The sample-local provider is single-instance: one configured tenant and agent instance, plus a separate blueprint secret. It has no managed-identity option and requests tokens from login.microsoftonline.com, so sovereign clouds need provider changes. Multi-instance or multi-tenant deployments should cache per agent/tenant and reuse the hosting connection credential.

Accepted OBS tokens are app-only tokens with idtyp=app, valid nonempty roles, or absent idtyp with nonempty oid == sub; any scp claim is rejected. See Agent 365 observability S2S export for route details and validation commands.

Working with User Identity

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

Field Description
activity.from_property.id Channel-specific user ID (e.g., 29:1AbcXyz... in Teams)
activity.from_property.name Display name as known to the channel
activity.from_property.aad_object_id 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 task instructions for personalized responses via the {user_name} input variable.

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 on_installation_update in host_agent_server.py:

Action Description
add Agent was installed — send a welcome message
remove Agent was uninstalled — send a farewell message 
if action == "add":
    await context.send_activity("Thank you for hiring me! Looking forward to assisting you in your professional journey!")
elif action == "remove":
    await context.send_activity("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. This is the recommended pattern for agentic identities in Teams.

Important: Streaming (SSE) is not supported for agentic identities in Teams. The SDK detects agentic identity and buffers streaming into a single message. Instead, call send_activity multiple times to send multiple messages.

Pattern

  1. Send an immediate acknowledgment so the user knows work has started
  2. Run a typing indicator loop — each indicator times out after ~5 seconds, so re-send every ~4 seconds
  3. Do your LLM work, then send the response

Typing Indicators

  • Typing indicators show a progress animation in Teams
  • They have a built-in ~5-second visual timeout
  • For long-running operations, re-send the typing indicator in a loop every ~4 seconds
  • Typing indicators are only visible in 1:1 chats and small group chats (not channels)

Code Example

# Multiple messages: send an immediate ack before the LLM work begins.
# Each send_activity call produces a discrete Teams message.
await context.send_activity("Got it — working on it…")

# Send typing indicator immediately (awaited so it arrives before the LLM call starts).
await context.send_activity(Activity(type="typing"))

# Background loop refreshes the "..." animation every ~4s (it times out after ~5s).
# asyncio.create_task is used because all aiohttp handlers share the same event loop.
async def _typing_loop():
    while True:
        try:
            await asyncio.sleep(4)
            await context.send_activity(Activity(type="typing"))
        except asyncio.CancelledError:
            break

typing_task = asyncio.create_task(_typing_loop())
try:
    response = await agent.invoke(user_message)
    await context.send_activity(response)
finally:
    typing_task.cancel()
    try:
        await typing_task
    except asyncio.CancelledError:
        pass

Running the Agent

Option 1: Run via Agent Runner (Standalone)

uv run python -m crew_agent.agent_runner "London"
# or
uv run agent_runner "San Francisco, CA"

This option integrates with Microsoft 365 Agents SDK for full observability and MCP tooling.

  1. Ensure .env is configured (see above)

  2. Get a bearer token for MCP authentication:

    a365 develop get-token -o raw

    Copy the token to BEARER_TOKEN in .env

  3. Start the agent host:

    uv run python start_with_generic_host.py

  4. Test endpoints:

    • Health check: http://localhost:3978/api/health
    • Bot Framework: http://localhost:3978/api/messages

  5. Connect with Agents Playground:

    agentsplayground -e "http://localhost:3978/api/messages" -c "emulator"

Example Prompts

Find weather in Dublin and email the information to user@example.com under the subject "Weather Report"

What's the weather in Seattle? Is it safe to drive with summer tires?

Architecture

Agent Workflow

┌─────────────────────────┐     ┌──────────────────────────┐
│  Weather Information    │────▶│  Driving Safety          │
│  Specialist             │     │  Specialist              │
│  (Tavily + MCP Tools)   │     │  (MCP Tools only)        │
└─────────────────────────┘     └──────────────────────────┘
         │                                │
         ▼                                ▼
   Weather Report              Safety Assessment + Email

MCP Tools Available

When connected with a valid bearer token, the following MCP servers are available:

Server Tools Description
mcp_MailTools 20 tools Send/receive emails, manage drafts, attachments
mcp_CalendarTools 12 tools Create/manage events, check availability
mcp_M365Copilot 1 tool Query Microsoft 365 Copilot

Observability Scopes

All agent operations are traced with Agent 365 observability:

  • InvokeAgentScope: Wraps the entire agent invocation
  • InferenceScope: Wraps LLM calls (CrewAI crew execution)
  • ExecuteToolScope: Wraps each MCP tool execution

File Structure

sample_agent/
├── .env.template              # Environment template with documentation
├── start_with_generic_host.py # Main entry point
├── host_agent_server.py       # Agent 365 SDK hosting + observability setup
├── agent.py                   # CrewAI agent wrapper with observability
├── turn_context_utils.py      # Turn context utilities
├── ToolingManifest.json       # MCP server definitions
└── src/crew_agent/
    ├── config/
    │   ├── agents.yaml        # Agent definitions
    │   └── tasks.yaml         # Task definitions
    ├── crew.py                # CrewAI crew orchestration
    └── tools/
        └── custom_tool.py     # WeatherTool (Tavily)

Troubleshooting

Common Issues

Issue Solution
No MCP tools discovered Ensure BEARER_TOKEN is set and not expired. Run a365 develop get-token -o raw
Weather tool returns no data Set TAVILY_API_KEY in .env
Azure OpenAI 401 error Verify AZURE_API_KEY and AZURE_API_BASE are correct
Duplicate emails sent Update tasks.yaml - only Driving Safety agent should send emails
CrewAI tracing 401 This is expected when CrewAI cloud tracing is not configured. It can be safely ignored; local observability still works

Logs to Check

INFO:agent:✅ 33 MCP tool(s) available:        # MCP connected successfully
INFO:agent:📊 Created observable wrapper       # Tool wrappers with ExecuteToolScope
INFO:agent:🔧 Calling MCP tool: SendEmail...   # Tool execution

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.