Bring your agents aboard

Coding agents do real work now, and they leave real messes: servers you never knew started, processes still alive an hour after the session ended. Pier fixes this at the root. Give your agent Pier's tools, and everything it runs happens somewhere you can see.

Agents work through Pier

Connect an agent (Claude Code, or anything that speaks MCP) and it works through Pier itself, on the same surface you use: starting dev servers, reading logs, installing dependencies, running package-manager and project CLI commands, standing up local services like Postgres or Redis, and creating share links.

You decide how much of that it may touch. A Permissions panel gives you a switch per capability, so you can let an agent read your projects and run servers while keeping services or public links off limits. Turning one off does not break your agent: Pier tells it the capability is off, so it asks you or falls back to its own shell.

Every tool call is logged, so "what did it just do?" always has an answer, and a separate durable record keeps everything an agent actually changed. A server it starts shows up exactly like one you started: live logs, the port, a real local domain, a stop button. And when an agent wants to put something on the internet, or run something destructive like resetting a database, Pier asks you first, every time. Nothing consequential happens without you.

Nothing left running

Agents forget to clean up; it's in their nature. Pier tracks what each agent session started, and when the session ends, it surfaces whatever was left behind so you can reclaim it in a click. No orphaned dev server burning a port overnight.

Calm, still

No red badges, no urgent anything. Pier shows which agents are connected, retires the ones that leave, and keeps the whole thing feeling like a tidy desk, agents included.