A developer tired of context-switching built a script that lets Claude answer Microsoft Teams messages on their behalf - and the approach is almost comically simple.
Instead of wrestling with Microsoft's Graph API, Azure AD permissions, and webhook configurations, the project uses a batch file (or shell script on Mac/Linux) that runs claude -p in a loop. Claude reads incoming Teams messages and generates responses with full access to the developer's local repositories. When a coworker asks about code architecture or a specific project, Claude can give answers grounded in the actual codebase rather than generic responses.
The appeal is obvious for anyone who spends half their day answering "how does this work?" questions in chat. The script runs from a terminal, stays local, and avoids the usual enterprise API overhead that kills most automation projects before they start.
There are clear limitations. This isn't an official integration - it's a clever workaround that reads the Teams interface directly rather than using Microsoft's approved APIs. That means it could break with any Teams UI update, and it's not something you'd deploy across an organization. It also raises questions about message privacy and whether your company's IT policy allows automated responses from third-party AI tools.
But as a personal productivity hack for developers who are drowning in internal chat, it's a neat proof of concept. The pattern of piping workplace messages through a local AI with repo access is something we'll likely see proper integrations handle soon - Anthropic's Claude already supports MCP (Model Context Protocol) for connecting to external tools, and Microsoft has been building Copilot deeper into Teams. The gap between this DIY approach and an official solution is shrinking fast.