ClawHosters
by Daniel Samer
6 min read
Microsoft 365 Copilot costs $30 per user per month. For a 50-person team, that's $18,000 a year before anyone sends a single prompt.
There's another way. OpenClaw connects your own LLM (Claude, GPT-4, Gemini, whatever you prefer) directly to Microsoft Teams. You control the model, you control the data, and you pay only for the tokens you actually use. According to SpeakWise's Teams statistics, 360 million people use Teams every month and 93% of Fortune 100 companies rely on it. Your AI agent should probably live where your team already works.
What Changed in v2026.3.24
The v2026.3.24 release is a big one for Teams users. OpenClaw migrated from the deprecated Bot Framework SDK (which Microsoft officially retired on December 31, 2025) to the new Teams AI Library and Teams SDK.
What that means in practice:
Streaming replies in 1:1 chats. The bot types in real time instead of dumping a wall of text after thinking for 20 seconds.
Welcome cards with prompt starters. New users see suggested prompts when they first message the bot. Reduces the "what do I even ask?" problem.
Typing indicators. The bot shows the "typing..." bubble while generating a response, so people know it's working.
Native AI labeling. Teams automatically marks bot responses as AI-generated. No extra work on your end.
If you're on an older OpenClaw version, you'll want to update before attempting the Teams integration. The old Bot Framework path is gone.
What You Need
The setup requires four things:
- An Azure account. Free tier works fine. The Azure Bot Service standard channel tier is free. You're not paying Microsoft for this.
- OpenClaw v2026.3.24 or newer. The Teams plugin won't work with older versions.
- The @openclaw/msteams plugin. Since the January 15, 2026 breaking change, the Teams connector ships as a separate package. Install it with
npm install @openclaw/msteams. - A public HTTPS endpoint. Teams needs to reach your OpenClaw instance over the internet. More on this in a minute.
Setup Overview
I won't walk through every click (the OpenClaw Teams docs do that well), but here's the flow so you know what you're getting into.
Step 1: Install the plugin. Run npm install @openclaw/msteams in your OpenClaw directory.
Step 2: Create an Azure Bot resource. In the Azure Portal, create a new Azure Bot. This gives you three values: appId, appPassword, and tenantId. Copy them.
Step 3: Configure OpenClaw. Add a msteams block to your openclaw.json with those three values. Set replyStyle to either "thread" or "top-level" depending on how you want bot responses to appear in channels.
Step 4: Build and upload the Teams app manifest. OpenClaw generates a Teams app package (a .zip with manifest.json, two icons). Upload it through the Teams Admin Center or sideload it for testing.
Step 5: Approve users. This is where most people get stuck.
The #1 Gotcha: Nobody's Talking to Your Bot
You installed everything. You message the bot in Teams. Nothing happens. No response, no error, just silence.
The default dmPolicy in OpenClaw is "pairing". That means the bot will not respond to any user until you explicitly approve them. It's a security feature (you probably don't want every person in your org burning through your API credits on day one), but it catches everyone off guard.
Fix it by running openclaw devices approve for each user, or set dmPolicy to "open" if you trust your org. In my experience, starting with "pairing" and approving users one by one is the safer move. You can always switch to "open" later.
Why Teams Is Harder to Self-Host Than Telegram or Slack
Here's something that surprised me. Telegram and Slack integrations work because OpenClaw connects outward to their APIs. Your server doesn't need to be publicly reachable. Teams works the opposite way. Microsoft's servers need to call your OpenClaw instance via a public HTTPS endpoint.
If you're self-hosting, that means setting up a persistent public URL. Most people reach for ngrok or Cloudflare Tunnel, but those are fragile for production use. Ngrok free URLs change on restart. Tunnels drop connections.
With ClawHosters, every instance gets a permanent HTTPS endpoint from day one. No tunnel setup, no DNS juggling, no certificate renewals. Your Teams bot just works at https://your-subdomain.clawhosters.com. And since ClawHosters auto-updates your OpenClaw instance, you were already on v2026.3.24 before this blog post existed.
If you want to skip the infrastructure headaches entirely, deploy your first agent and have Teams connected in under an hour. Plans start at $19/month.
One More Thing: replyStyle
Teams doesn't tell the bot whether a channel uses threaded or flat conversation mode through the API. You have to set replyStyle manually in your config. Pick "thread" if your team uses reply threads, "top-level" if they don't. Get this wrong and bot responses show up in confusing places. There's no auto-detection. Just check how your team actually uses channels and match it.
