cramt/m365-copilot-proxy/skill

M365 Copilot Provider

This skill configures OpenClaw to use Microsoft 365 Copilot as the LLM backend through an OpenAI-compatible proxy server.

How it works

The M365 Copilot proxy translates OpenAI-format chat completion requests into M365 Copilot WebSocket calls. It supports:

• All M365 Copilot models: m365-copilot, quick, think-deeper, gpt-5.4, gpt-5.3, gpt-5.2 (with quick/reasoning variants)
• Tool calling: Automatic translation of OpenAI tool_call format through prompt injection + fenced code block parsing
• Agent mode: Uses a Copilot Studio agent for server-side system prompt control
• Streaming: Full SSE streaming support
• Session reuse: Conversations share a single M365 session, saving quota (600 msg limit per conversation)
• Delta messages: Follow-up turns only send new messages, not the full history

Setup

Run m365-openclaw-setup to configure OpenClaw. Add --start to also launch the proxy:

# Configure only
m365-openclaw-setup
 
# Configure and start proxy
m365-openclaw-setup --start
 
# Configure with custom port
m365-openclaw-setup --start 8080

Authentication

Auth credentials are stored in ~/.config/opencode-m365/secrets.json:

• email: Your M365 email address
• password: Your M365 password
• mfaSecret: Your TOTP MFA secret (base32)

The proxy handles token refresh automatically using MSAL with PKCE.