Microsoft Teams Bridge for OpenClaw

@komaa/openclaw-msteams-bridge is a Microsoft Teams voice and video agent (CVI) for OpenClaw. It turns an ordinary Teams call into a true two-way video conversation: the agent sees what you show it, talks back in real time, and appears in the call as an animated, lip-synced avatar.

It is one plugin, depending only on the published openclaw plugin-sdk + api.runtime. No fork, no vendored runtime, no trusted-plugin privileges required.

Three pillars

Details of each capability follow below.

Capabilities

• Two dialogue modes - realtime speech-to-speech (OpenAI or Azure OpenAI), or a streaming STT to agent to TTS pipeline that works with any provider.
• Vision - reads camera and screen-share frames, keeps a continuous ambient view, retains a retroactive keyframe history, and stays inside a per-call budget cap.
• Group and meeting etiquette - stays silent until addressed by a wake phrase, then answers through a short follow-up window; 1:1 calls always answer; every frame and utterance is attributed to the right speaker.
• Outbound call-backs - places a call, speaks the result, and hangs up, with a voicemail fallback when no one answers.
• Meeting recap and minutes - end-of-call summary of key points, decisions, and action items, plus an on-demand .docx of minutes with per-person attribution.
• Avatar driver cues - expression changes, viseme lip-sync, and picture-in-picture image sharing, rendered by the StandIn bridge.
• Chat governance - an "Ask about this" message action, voice-message transcription, an audit-log mirror, and outbound DLP redaction.
• Secure transport - a replay-proof HMAC handshake, a caller allowlist that is closed by default, and a recording-status gate that holds media until recording is active.

Getting started

This plugin adds voice and video (CVI) on top of OpenClaw's Microsoft Teams chat channel, so set those up first:

1. Install OpenClaw using the official docs at docs.openclaw.ai.

2. Add Microsoft Teams as a channel (bot app + credentials) following the OpenClaw Teams channel docs.

3. Use the StandIn sandbox (standin.komaa.com/sandbox, free, no Teams bot needed), the hosted media bridge that joins the call and connects to this plugin. Add your own Teams bot later at standin.komaa.com for inbound calls.

4. Add this plugin. The one-line installer detects your OpenClaw install and walks you through the config (mode, shared secret, provider key), then prints the next steps:

   curl -fsSL https://standin.komaa.com/install.sh | bash
   

   Prefer to do it by hand? See Install and Configuration below.

Requirements

• An OpenClaw install (host >= 2026.6.10).
• StandIn to bridge the call: start free in the sandbox (no Teams bot), or add your own bot at standin.komaa.com for inbound calls. It is the hosted media bridge that joins the Teams call and connects to this plugin's WebSocket.
• For realtime mode: a realtime voice provider + key (OpenAI or Azure OpenAI). For streaming mode: your OpenClaw-configured STT / TTS / agent (no realtime key needed).

Install

Do these in order. Steps 1 and 2 are prerequisites (see Getting started above); this plugin is step 3.

1. OpenClaw is installed and running (host >= 2026.6.10).

2. Microsoft Teams is added as a channel (bot app + credentials).

3. Install this plugin, then restart the gateway so it loads:

   openclaw plugins install npm:@komaa/openclaw-msteams-bridge
   openclaw gateway restart
   

4. Configure it under plugins.entries."msteams-voice".config - at minimum set sharedSecret (to match StandIn), inboundPolicy, and your provider key. See Configuration and Security below. The server refuses to start until sharedSecret is set.

5. Connect StandIn to the plugin's WebSocket (start in the sandbox) and place a test call.

Prefer a guided setup? The one-line installer detects your OpenClaw install and walks you through steps 3 and 4 (mode, shared secret, provider key), applying the secure defaults for you:

curl -fsSL https://standin.komaa.com/install.sh | bash

Also on ClawHub: openclaw plugins install clawhub:@komaa/openclaw-msteams-bridge (OpenClaw falls back to npm automatically if the ClawHub fetch fails). The package ships prebuilt (v0.1.10+): no build step either way.

Two modes

Set mode to "realtime" or "streaming". If omitted, the runtime auto-selects realtime when a realtime provider resolves, else streaming.

Configuration

Config lives under plugins.entries."msteams-voice".config. sharedSecret must match the value set in your StandIn dashboard. Set bindAddress to 0.0.0.0 so the hosted bridge can reach it.

Realtime (OpenAI)

{
  "plugins": {
    "entries": {
      "msteams-voice": {
        "config": {
          "enabled": true,
          "mode": "realtime",
          "bindAddress": "0.0.0.0",
          "port": 9442,
          "path": "/voice/msteams/stream",
          "sharedSecret": "<same secret as in StandIn>",
          "requireRecordingStatus": true,
          "inboundPolicy": "allowlist",
          "allowFrom": ["<caller AAD object id>"],
          "realtime": {
            "provider": "openai",
            "providers": {
              "openai": { "apiKey": "<key>", "model": "gpt-realtime" }
            }
          }
        }
      }
    }
  }
}

Realtime (Azure OpenAI)

Azure is the openai provider plus azureEndpoint and azureDeployment:

"realtime": {
  "provider": "openai",
  "providers": {
    "openai": {
      "apiKey": "<azure-key>",
      "azureEndpoint": "https://<resource>.cognitiveservices.azure.com",
      "azureDeployment": "gpt-realtime"
    }
  }
}

Streaming (no realtime key)

"mode": "streaming",
"stt": {
  "provider": "<your-stt-provider>",
  "providers": { "<your-stt-provider>": { "apiKey": "<key>" } }
}

In streaming mode the TTS and agent come from your OpenClaw configuration. STT uses stt.provider if set, else your configured transcription provider, else a VAD-segmented file fallback. Group-call gating, DTMF, and vision all work in streaming mode too.

Outbound call-backs (optional)

"outbound": {
  "enabled": true,
  "workerBaseUrl": "https://<your-standin-endpoint>",
  "tenantId": "<aad-tenant-id>",
  "answerTimeoutMs": 120000,
  "defaultMode": "notify"
}

Security

The plugin ships with secure defaults, and the recommendation is simple: keep them. Each option below states its safe default and why it is safe. You only relax a default when your deployment genuinely needs it, and only in the narrow way described.

In short: set a strong sharedSecret, keep inboundPolicy restrictive with an explicit allowFrom, leave requireRecordingStatus on, keep bindAddress as tight as your topology allows, and only raise toolPolicy for callers you trust.

Key reference

Full reference in the Configuration Reference. Common keys:

Core

Group call

Realtime

Streaming

Outbound

Links

• Plugin docs: komaa-com.github.io/openclaw-msteams-bridge
• StandIn (hosted service) docs: docs.komaa.com
• Source: github.com/komaa-com/openclaw-msteams-bridge
• npm: @komaa/openclaw-msteams-bridge

_{Built by Komaa.com - MIT licensed}