Download Log in Sign up

Browz for AI agents

A separate, secure browser for every agent task.

Let your AI agent browse in Browz instead of in your own browser, whether it runs on Claude, OpenAI, Gemini or Grok models, or on your own framework. Each task gets its own isolated profile with a real device fingerprint and its own proxy, sessions persist between runs, and you can step in whenever a site asks for a person.

Set up MCP Download Browz

Why agents need their own browser

AI agents are browsing.

More of the web is visited by software every year, and more of that software is AI agents acting for people. An agent signed in to your own browser can reach every account you have; one that a page tricks can act on all of them. Each task belongs in a browser of its own.

40%

of enterprise apps will feature task-specific AI agents by the end of 2026, up from less than 5% in 2025 (a forecast).

Source: Gartner, 26 Aug 2025
79%

of senior US executives say AI agents are already being adopted in their companies.

Source: PwC AI Agent Survey, May 2025
53%+

of all web traffic in 2025 was automated. In June 2026 bots passed humans on Cloudflare’s network, at 57.5% of requests for web pages, a year ahead of its CEO’s 2027 forecast.

Sources: Imperva Bad Bot Report, 29 Apr 2026; Matthew Prince, Cloudflare CEO, 3 Jun 2026; Cloudflare Radar, via DROIDS!, 4 Jun 2026
23.6%

of prompt-injection attacks succeeded against a browsing agent in its maker’s own tests without defences; 11.2% with them.

Source: Anthropic, 25 Aug 2025

Gartner has advised security teams to block AI browsers until the risks are assessed. Source: Gartner, as reported by The Register, 8 Dec 2025. The Model Context Protocol, which Browz supports, passed 97 million monthly SDK downloads (Model Context Protocol, 9 Dec 2025). Figures are quoted as each publisher states them.

Available now

What your agent gets today.

01

An isolated profile for every agent task.

Your agent creates a profile for the job, starts it, works in it and deletes it when it is done. Each profile is a separate browser with its own cookies, storage, device fingerprint and proxy, so one task never sees another’s accounts, and none of them sees your own browser.

  • Create, start, stop and delete profiles from the agent
  • Its own fingerprint and checked proxy exit
  • Nothing shared with your everyday browser
02

Sessions that persist.

Sign a profile in once. When the agent comes back tomorrow, the cookies and site storage are still there, so it carries on signed in instead of logging in again on every run. Sessions survive app restarts and crashes, and sync end-to-end encrypted to your other computers.

  • Cookies and storage kept per profile
  • Survives restarts and crashes
  • Synced end-to-end encrypted
03

A person can step in.

Profiles open as ordinary browser windows on your computer. When a site asks for a password, a 2FA code or a CAPTCHA, you complete it yourself in that window, and the agent continues in the same signed-in session. A built-in hand-off, where the agent pauses and asks you, is coming.

  • Visible windows you can take over
  • Same session before and after
  • Guided hand-off for logins, 2FA and CAPTCHAs Coming
04

Access that stays on your computer.

The MCP server talks to the Browz API on 127.0.0.1 only, behind a token you can rotate from the app. Lock a profile behind its own password so an agent can’t open it, block sites per profile, and read the local action log of what was created, started, stopped and deleted.

  • Loopback-only API with a rotating token
  • Profile locks and per-profile blocklists
  • Local action log
  • Per-agent scopes, consent prompts and a full audit trail Coming

MCP setup

Connect an agent in three steps.

The Browz app ships with an MCP server. It runs on your computer next to the app.

  1. Install Browz and sign in

    Browz has to be open: the MCP server works through the app’s own local API, with its token.

  2. Add Browz to your MCP client

    Paste the config below into any client that runs a local MCP server over stdio, such as Claude Desktop, Claude Code, Codex CLI, Gemini CLI, Cursor or VS Code, or into your agent framework’s MCP settings.

  3. Ask your agent

    For example: “Create a Browz profile called Supplier check, open example.com in it and take a screenshot.”

Setup for each AI client Developer quickstartAutomation and API

Coming

What we are building next.

These are in development and not in the app yet. Each one will be listed under Available now when it ships.

01

A remote MCP server

Connect hosted agents to your Browz from anywhere, signing in with OAuth and granting each one only the access it needs.

Coming
02

Page snapshots for agents

A compact view of each page with stable references to its elements, so agents act on pages with fewer tokens and fewer misclicks.

Coming
03

Throwaway and pooled profiles

Profiles made for one task and removed after it, pools of ready profiles, and resuming a task where it stopped.

Coming
04

Hand-off to a person

The agent pauses at a login, 2FA code or CAPTCHA, tells you, and resumes when you are done.

Coming
05

Consent and audit

Approve sensitive actions before they happen and review everything each agent did, per profile.

Coming
06

A benchmark

Published results for agents running in Browz.

Coming

FAQ

About agents in Browz.

Which AI agents work with Browz?

Any agent or client that can run a local MCP server over stdio, such as Claude Desktop, Claude Code, Codex CLI, Gemini CLI, Cursor and VS Code, and agent frameworks with MCP support, whichever model they use. Agents built with Playwright or Puppeteer can also connect to a running profile over the Chrome DevTools Protocol. Agents that run in the cloud, such as chat apps on the web, need the remote MCP server, which is coming.

Does the agent see my own browser?

No. It works only in Browz profiles, and each profile is a separate browser with its own cookies and storage. Your everyday browser, its passwords and its sessions are never part of it.

What happens at a login, 2FA code or CAPTCHA?

Today you complete it yourself in the profile’s window, and the agent carries on in the same session. A guided hand-off, where the agent pauses and asks you, is coming.

Can I limit what an agent may do?

Today the agent has the access of your local API token: rotate it to cut every client off at once, lock profiles it should not open, and block sites per profile. Scoped access for each agent, with consent prompts and a full audit trail, is coming.

Is it allowed to use agents on any site?

Use agents only where the site’s terms and the law allow it, and on accounts you are entitled to use. Our acceptable use policy applies to agents the same as to people.

Get started

Your agent. Its own browser.

Download Browz See pricing

For Mac, Windows and Linux. Free for 5 profiles.