Now You Can Talk to Claude in the Middle of the Task

#ai #opensource #cluade #mcp

You know that moment?

Claude Code is mid-refactor on a massive task. It hits a decision point and needs your input. But you're grabbing coffee with your friend, and it's taking way longer than the 5 minutes you promised yourself.

You could have handled it all right there over coffee and still enjoyed your break.

I got tired of that loop, so I built claude-slack-bridge: an MCP server that creates a real-time, two-way communication channel between Claude Code and Slack.

And honestly? It changed the way I think about human-in-the-loop AI workflows.

How It Works

The bridge operates in two directions:

Direction 1: Claude → You
Claude pauses mid-task, sends you a question on Slack, waits for your reply, and picks up exactly where it left off.

Claude: "Should I refactor the auth module to use JWT 
        or keep session-based auth? The migration would 
        touch 12 files."

You (from your phone): "Go with JWT. Skip the legacy endpoints."

Claude: *continues working*

Direction 2: You → Claude
You write a message in the Slack channel. Claude Code picks it up, executes it, and replies in the same thread — with full context and session awareness.
It's like texting your AI teammate.

Why Not Just Use Official Slack MCP?

Fair question. Here's the key difference:
Slack MCP is designed to make Slack a data source — so Claude can search channels and pull context from conversations.

claude-slack-bridge does the opposite. It uses Slack as a remote control for your local terminal session. It's built specifically for the human-in-the-loop pattern where your CLI agent needs your approval or guidance to proceed.

Think of it this way:

Official Slack MCP = Claude reads your Slack
claude-slack-bridge = Claude talks to you through Slack

The Technical Bits (For the Curious)

⚡ Zero Polling - The bridge uses OS-level blocking — no busy-wait loops, no webhook servers to manage. When Claude asks a question, the process sleeps at the kernel level until your reply arrives.
🐳 Single Docker Container - One container runs in the background and serves all your projects. No per-project setup, no port conflicts.
🔀 Parallel Sessions via Unix Domain Sockets - Running multiple Claude Code sessions? Each one gets routed to the correct Slack thread automatically.
🧠 Session Resume - Container restarts? No problem. Claude reconnects to the existing Slack thread history and picks up context. The conversation thread is the persistence layer.

Getting Started

The setup is straightforward:

Clone the repo
Configure your Slack bot token and channel
Run the Docker container
Add the MCP server to your Claude Code config

Full instructions are in the README.

The Repo
The project is open source under MIT license:
🔗 github.com/tomeraitz/claude-slack-bridge.
If this sounds useful to you, I'd love a ⭐ on the repo — it really helps with visibility.

I write about AI agent workflows, Claude Code, and building developer tools. Follow me here or check out vibki.com for more.