ClawHosters
by Daniel Samer
5 min read
320 million people use Microsoft Teams every month. That's where decisions happen, questions get asked, and work actually moves forward. So it makes sense to put your AI assistant there too.
Microsoft wants you to pay $30 per user per month for Copilot. For a 50-person team, that's $18,000 a year. OpenClaw gives you a self-hosted alternative that connects to Teams with your own LLM keys. You pick the model. You control the data. And the per-user cost drops to basically nothing.
What You Need Before Starting
The setup takes about 20 minutes if you have these ready:
An Azure account (the free tier works fine for bot registration)
OpenClaw v2026.3.24 or newer (the March 2026 release migrated to the official Microsoft Teams SDK)
A public HTTPS URL pointing to your OpenClaw instance
If you're running on ClawHosters, you already have the HTTPS URL. Every instance gets a *.clawhosters.com subdomain with TLS certificates handled automatically. No ngrok tunnels, no reverse proxy headaches.
Azure Bot Registration
This is the part most people get stuck on. Microsoft changed their bot registration process in mid-2025, and a lot of older tutorials are wrong now.
Head to the Azure Portal and create a new Azure Bot resource. Pick Single Tenant for the app type. Multi-tenant bots were deprecated on July 31, 2025, so new registrations must be single-tenant.
You'll get three values from this process:
App ID (sometimes called Client ID)
App Password (a client secret you generate)
Tenant ID (your Azure AD directory ID)
Write these down. You'll need all three in a minute.
Under the bot's Channels settings, add the Microsoft Teams channel. Then configure the Messaging Endpoint to point at your OpenClaw webhook URL. The format is https://your-instance.clawhosters.com/webhooks/msteams.
Installing the Teams Plugin
The Teams connector moved out of OpenClaw core on January 15, 2026. It's now a separate plugin. SSH into your instance (or use the ClawHosters dashboard) and run:
openclaw plugins install @openclaw/msteams
Then add your Azure credentials to the environment config:
MSTEAMS_APP_ID: "your-app-id-here"
MSTEAMS_APP_PASSWORD: "your-app-password-here"
MSTEAMS_TENANT_ID: "your-tenant-id-here"
On ClawHosters, you can paste these directly into the dashboard under Connectors > Microsoft Teams. No SSH required.
Restart OpenClaw, and the bot should come online in Teams within a couple of minutes.
What Your Microsoft Teams Chatbot Can Do
Once connected, your microsoft teams ai assistant handles a few things that Copilot honestly doesn't do well:
Streaming replies in DMs. When someone messages the bot directly, responses stream in real time. There's a hard 15-second webhook timeout on the Teams platform and a 2-minute max stream duration, but for most questions that's plenty. In group channels, the bot sends the full response at once instead of streaming. That's a Teams platform limitation, not an OpenClaw one.
Welcome cards. New users get an adaptive card explaining what the bot can do and how to interact with it. You can customize the card content in your OpenClaw config.
AI labeling and typing indicators. Teams shows a typing indicator while the bot processes, and responses are automatically labeled as AI-generated. Microsoft started requiring this in late 2025.
Your choice of LLM. Connect GPT-4o, Claude, Gemini, Llama, or any OpenAI-compatible API. The Teams connector is just the interface. The brain behind it is whatever model you configure in OpenClaw.
Security Settings You Should Know About
The defaults are strict on purpose. I think that's the right call for a corporate Teams environment.
DM policy is set to pairing by default. This means unknown senders get ignored. Only users who have been paired (verified) can chat with the bot in DMs. You can change this to open if you want anyone in your tenant to use it immediately, but think carefully about that.
Group policy is allowlist by default. The bot only responds in group chats where it's been explicitly added, and even then, it requires an @mention before responding (the requireMention flag is true).
User allowlists use AAD object IDs, not display names. Display names aren't unique in Azure AD, so relying on them would be a security risk. Grab object IDs from the Azure Portal under Users > Properties.
