For Developers

The automation you'd never bother to write

Aera is a browser with an agent inside. Describe a chore in one sentence, like "run the signup flow every hour and tell me when it breaks", and it runs on a schedule, on your machine, with a step-by-step log. Free to download, and free to run on a local model you host. Sign-in required. Works standalone or wired into Cursor and Claude Code.

Download for Free

Why this exists

You've already tried to script this.

Puppeteer dies on a renamed div. Headless browsers die at the login wall. And the checks that waste the most time are the ones too small to ever justify a script.

Really signed in

Checks run in a real Chromium window with your existing sessions. No login wall, no headless profile to re-auth, nothing to host.

Described, not scripted

One sentence replaces the script. When a page changes, the agent re-reads it the way you would, instead of dying on a selector.

It fails loudly

Agent runs aren't deterministic, and we don't pretend otherwise. Every run logs each step; when one breaks, it stops and shows you the log instead of guessing.

From your IDE

Cursor and Claude Code can drive it too.

Aera ships a local MCP server, so your coding agent gets a real browser mid-task: pull live docs, check a dashboard, diff a changelog, and hand structured results back.

Schedules live in Aera; create and inspect them from any MCP client, or skip the IDE entirely
Step-by-step run history in the Aera sidebar
Schedule recurring checks: QA flows, docs diffs, partner portals

MCP · triggered from Cursor

Pull the latest webhook docs for the payments API and diff against what we implemented.

Completed 8 steps in 13s

Navigated to the payments API changelog

Read webhook signature requirements

Compared against the cached spec from last sprint

Diff ready · 2 changes

Signature header renamed; your verify step breaks

Retry window extended from 24h to 72h

Quickstart

Wired into your IDE in under a minute.

Point Cursor, Claude Code, or any MCP client at one stdio command. If Aera isn't running, the tool call starts it; if the browser is closed, it wakes.

Install Aera and finish first-run once
Paste the endpoint into your MCP config
Ask your agent to run your signup flow and say where it fails

.cursor/mcp.json

{
  "mcpServers": {
    "aera-browser": {
      "command": "aera_mcp_proxy",
      "args": ["--stdio"]
    }
  }
}

Claude Code:claude mcp add aera-browser -- aera_mcp_proxy --stdio

Some of what your agent getsnavigate_to_urlanalyze_tabinteractrun_javascriptread_console_logsview_network_requestscapture_tab_screenshotcreate_scheduled_task

Why you'd actually let it in

Your credentials never enter the loop.

Aera drives sessions that already live in your browser, on your machine. Your passwords are never sent anywhere: no credential vault on our servers, no cloud replay of your logins. What the model does read, the pages you point it at, goes to the provider serving the model you picked, never to one that trains on it. Or run a local model, and then the page never leaves your machine.

No credential handoff

The agent acts inside your signed-in sessions instead of holding your passwords. None of its tools touch the password store. That access is real power, which is exactly why every action it takes is logged.

Local-first storage

Chats, tasks, files, and run history live in SQLite on your disk. Uninstall and it's gone. And if we vanished tomorrow, the browser and every file it wrote would still be yours.

Falsifiable on purpose

Point it at a local model and watch the network interface: your prompt and the page go to your own endpoint, not to us. Aera still talks to our server about your account, and the security page lists what that carries.

Early-stage honesty: local stdio connectors currently run unsandboxed, and OS-vault storage for connector secrets is in progress. The security page tracks exactly what's shipped. Read the security model

Before you trust it

Don't take our word for any of this.

We are not going to hand you a case study. Every claim below is checkable on your own machine inside the first ten minutes.

verifiable

Runs with the browser closed. The scheduler is a separate service on your machine. Schedule a task, quit the browser, watch it fire anyway. Your machine does need to be awake: there is no hidden cloud runner.

Logs every step. Open any run and read each tool call it made. If it misbehaves, the log is the evidence.

Local SQLite. Your data is a file on your disk. Uninstall and it's gone; there is no server-side copy to leak.

No meter. Bring your own local model and it stays free. Hosted models are a flat $20 a month, not a per-run bill you discover later. An Aera account is required either way.

We ship weekly. Read the changelog →

Get started

Steal a starting point.

No API, behind SSO, too small to justify a sprint. Describe it once; it runs on cadence and raises an OS notification when something needs you.

Run the signup QA flow and notify me only when a step failshourly

Diff the partner API docs and flag breaking changesdaily

Check the status pages of every service we depend on15 min

Sweep our dependency changelogs and flag anything that breaks usweekly

Keep exploring: Docs · MCP in Aera · Security · Use cases