Claude Code and team chat: a practical guide
What changes when your coding agent can read your team's decisions, and how to set it up safely.

The short answer
Claude Code works best when it knows your team's context. Connect it to team chat over MCP and it can read channels, search past messages, open docs and create tasks, so you stop pasting context into prompts. In Dock you create a token on Connect AI, add one line to Claude Code, and the tools appear.
Setup in four steps
Create a token
Open Connect AI in Dock and create a personal access token.
Register the server
Run the `claude mcp add --transport http dock <url>/mcp --header "Authorization: Bearer <token>"` line the page prints for you.
Check it
Start claude and type /mcp; dock should be connected.
Ask
"Read #engineering since yesterday and list decisions and open questions."
Habits that keep it safe
Own bot for unattended work
Give it narrow access instead of your personal token.
Review before posting
Ask the agent to draft in busy channels.
Rotate tokens
Revoke from Connect AI when you stop using one.
Treat output as a draft
Agents can misread context.
What makes Dock different: it connects to your tools
Dock connects the places your team already works. Notion: connect your Notion pages, then type @notionpage_ in any chat to tag a page; everyone gets a preview card and your AI agents can read it. GitHub: link GitHub read-only, pin repositories to a chat, tag them with @github_ and ask the AI about them. Claude and Codex: use your own Claude or Codex plan inside Dock chats, or connect any MCP agent to your workspace with a token.
All of this sits next to chat, video meetings (with recording, transcripts and AI summaries), docs, tables, calendar, tasks and a drive, in a self-hosted app with no per-seat licence. You run the server yourself, and some integrations are young, so check the setup guides for what each one covers.
Questions
Does Claude Code need an API key for this?
No. MCP access uses a Dock token; Claude Code uses your Claude plan or API key as usual.
Can it create tasks?
Yes, with the task tools, for lists the token can reach.
Does it work with Codex and Cursor too?
Yes, any MCP client that supports HTTP servers and bearer tokens.