I got tired of half-baked M365 integrations so I built my own

I got tired of half-baked M365 "integrations" so I built my own. I've been using OpenClaw and wanted something pretty simple: let the agent work with Microsoft 365 reliably, without fighting a half-finished wrapper every time I asked it to do something useful. I tested a number of existing Outlook/M365 MCP implementations. Some covered email but barely anything else. Some had awkward tool design. Others felt more like demos than something I'd trust as part of a daily workflow. Eventually I stopped patching around them and wrote my own M365 skill. It now exposes 43 tools across: Outlook mail Calendar Contacts OneDrive Teams Tasks M365 MCP for OpenClaw The interesting part wasn't getting Microsoft Graph to answer requests. It was deciding what an AI agent should actually be allowed to do, how the tools should behave consistently, and where access should stop. For example, my main OpenClaw agent can use M365. My coding agent can't. Different job, different trust boundary. Funny how often "I'll just find an existing integration" turns into another software project. :) If anyone else needs it, you can find it on GitHub and ClawHub as m365-mcp.