MCPs

Plug your agent into the tools and data outside your project.

Out of the box, your agent sees exactly one thing: the files in your project. Your database, your issue tracker, your designs, the live internet, all invisible. An MCP is the connector that changes that, plugging the agent into one outside service so it can look things up and take actions there for real.

The name (Model Context Protocol) matters less than the analogy: it's USB for AI. One standard plug shape, and every service (GitHub, Supabase, Figma, your browser) can offer a socket. Connect one, and “check my database and find who signed up first” becomes something the agent actually goes and does.

Your agent

“Did anyone sign up while I was asleep?”

→

Supabase MCP

The connector. It offers the agent a menu of actions:

Query tablesInspect schemaRun SQL

→

Supabase

Your real database, your actual members.

The agent asks, the MCP does the work in Supabase, the answer comes back.

Connections our community actually uses:

Supabase

“Did that row save?” becomes a question you ask, not a dashboard you check.

Vercel

Check deployments and read build logs without leaving chat.

GitHub

Read and manage your code in the cloud.

Figma

Pull real designs in, so the agent builds from the source.

Linear

See and update the issues your team tracks.

A browser

Open your actual site and test how it behaves.

If you read the Integrations section, you already know what's under the hood. The MCP is making API calls on your behalf, key and all:

None of which you touch. Setup is one command. Here's Claude Code wiring itself to Supabase:

claude mcp add --transport http supabase https://mcp.supabase.com/mcp

One tip from our workshops: connect MCPs per-project rather than everywhere, and disconnect ones you're not using. Every active connection takes up standing room in the context window.

AI Clubhouse · Vibe Coding Guide