What is MCP, and why should a team care?
The Model Context Protocol in plain English: how AI agents get safe access to your team's chat, docs and tasks.
The short answer
MCP (Model Context Protocol) is an open standard that lets AI assistants call tools and read data from other apps through one common interface. Instead of pasting text into a chat window, you connect the assistant to a server that exposes what it may do, such as list channels, read messages, search docs or create a task.
For teams this matters because the useful context for an AI agent, the decisions, bugs and plans, lives in your team tools, not in the agent's window.
What an MCP connection can offer
Read
List channels, read messages, search, open docs, read meeting transcripts.
Write
Post a message or add a reaction where the connected account can.
Work
List task lists, create tasks, mark them complete.
How it works
A server exposes tools
Your team app runs an MCP server that lists the tools and what each does.
An agent connects
Claude Code, Codex CLI, Cursor, Windsurf and Gemini CLI can add the server with a URL and a token.
You ask in plain language
The agent picks the right tool, runs it and shows you the result.
MCP in Dock
Dock runs an MCP server on your own instance. You create a personal access token on the Connect AI page, register the server once in your agent, and the tools appear. The page prints the exact setup line for each agent with your URL filled in. See Connect Claude Code to your team chat.
Treat the token like a password: it carries your permissions. For anything unattended, use a bot account that only sees what you grant it.
Questions
Is MCP only for Claude?
No. It is an open standard that several agents and editors support.
Is it safe?
It is as safe as the access you give it. Limit tokens, prefer bots for unattended work and review before agents post.
Do I need to self-host to use MCP?
MCP works with hosted tools too. With Dock, the server runs on your own instance.