Until recently, connecting an AI assistant to a business tool meant a custom integration per pair — N tools × M assistants, built and maintained forever. The Model Context Protocol (MCP), introduced by Anthropic and since adopted across the industry, collapses that: a tool exposes its capabilities once, as an MCP server, and any MCP-capable assistant — Claude, ChatGPT, Cursor and the rest — can discover and use them.
The USB analogy, because it's accurate
Before USB, every peripheral shipped its own port and driver. USB didn't make printers better — it made connecting them a non-event, and the ecosystem exploded. MCP is that for AI: a server declares its tools (actions with typed inputs), the assistant calls them, OAuth handles who's allowed. Boring plumbing, deliberately — the 40%-of-enterprise-apps agent wave needs exactly this kind of boring.
What changes for people who run projects
Your tools stop being places you go and become capabilities your assistant has. The practical version, today: a PM in Claude asks "what's overdue across my projects?" or "client wants a loyalty screen — is that in the Atlas scope?" and gets a real answer, because their PM system exposes those exact tools over MCP — including the scope check with evidence attached.
Why we shipped one
Gavril's MCP server is live: project status, task search, team workload, document search and the scope guard, all callable from any MCP client your team already uses (details at gavril.me/auth.md). We think every serious business tool will do this within a couple of years, for the same reason everything eventually grew a USB port: once connecting is a non-event, not connecting is a decision against your users.
← All posts