An AI agent that actually uses your browser.

Auto Browser sits in your Chrome side panel. Describe a task in plain English — it sees the page, acts on your behalf, and reports back. No scripts, no servers, no vendor lock-in.

Install free Watch how it works
  • Works on any site — zero integration
  • Bring your own AI — 4 providers
  • Private by default — optional on-device
webmcp-sports.com
  • Chrome Manifest V3
  • Works on any site
  • Zero telemetry
  • On-device AI option
  • Keys stay local
How it works

Three moves, every task.

Auto Browser runs a perceive → plan → act loop. It keeps going until the task is done or it needs you. Every action is yours to approve.

  1. 01

    Ask

    Type what you want in plain English. Attach screenshots or audio if useful.

    “Find me a flight from NYC to Tokyo next Thursday, one-way.”
  2. 02

    See

    The agent reads the page's accessibility tree and captures a snapshot — it understands structure, not pixels.

    Snapshot taken → 47 elements, search form detected
  3. 03

    Act

    It clicks, types, scrolls, navigates, or runs the page's own WebMCP tools — always gated by your permissions.

    Fill form → submit → read results
Why Auto Browser

Four commitments, zero compromises.

Works everywhere

Any website, any time. No per-site integration, no API stubs, no 'connect your account' dance. The agent reads the page through standards the browser already speaks.

Bring your own AI

Four providers, one dropdown. Run Gemini Nano on-device for privacy, OpenRouter for any model, Google Gemini for frontier quality, or Ollama for air-gapped setups.

Learn more

Private by default

No analytics. No telemetry. Your API keys stay on your machine. Passwords and secret tokens are redacted before they can reach the model. Pick on-device AI and zero data leaves your browser.

Learn more

Safe by design

The agent asks before every write. Script execution always prompts. Redirect attacks are neutralized automatically. Banking, gov, and healthcare sites are protected by default.

Learn more
Use cases

Every workflow you keep putting off.

A generic, app-agnostic agent means the same install handles radically different jobs. Tap a card to watch it run.

Shop

Compare products and recommend one

The agent searches the store, reads specs and reviews on each option, and hands back a side-by-side summary with a clear recommendation.

“Compare all the soccer boots and suggest which one to buy.”
Extract

Pull structured data from any table

Complex HTML tables stop being complex. The agent writes and runs eval scripts (with your approval) to filter, sort, and extract exactly what you asked for.

“Give me the top 5 models cheaper than $2.”
Browse

Summarise what someone's posting

Point it at any X profile and ask what they're up to. The agent scrolls the timeline, reads the posts, and hands back a concise brief.

“What is Elon Musk talking about?”
Read

Summarise long articles

Drop it on a blog post, whitepaper, or release note. The agent reads the full page and returns a summary plus the key insights.

“Summarise this article and give me the key insights.”
Reserve

Make a restaurant reservation

It finds the reservation form, asks only for what it doesn't already know, and submits autonomously — you're in the loop only when it matters.

“Make a reservation for 2 next Friday, 7pm.”
Tickets

Book a movie ticket for Friday

It scans the showtimes on a cinema site, checks in on which film you want, and walks through seat-and-pay — pausing for your final approval.

“Let's see what movies are on and get a ticket for this Friday.”
AI providers

Four AIs. One dropdown. Zero lock-in.

Most agent tools marry you to a single model. Auto Browser lets you pick — and switch — without reconfiguring the agent or losing your conversation.

01

Chrome Built-in AI

Free. On-device. Private.

Gemini Nano runs locally inside Chrome. No API key, no network calls, no account. Downloaded once (~3–5 GB) and it works offline forever.

  • No API key
  • Runs offline
  • Zero data leaves your device
Best for Privacy-sensitive work · Offline use
02

Google Gemini

Frontier quality, direct API.

Point at Google's API directly. Great for complex reasoning tasks where latency and model quality both matter.

  • Direct Google API
  • Latest Gemini models
  • Your key, your quota
Best for Complex reasoning · Long context
03

OpenRouter

Every model, one key.

Claude, GPT, Llama, Mistral — any model in the OpenRouter catalog, switched with a dropdown. Image and audio support auto-detected per model.

  • Claude · GPT · Llama · more
  • Single API key
  • Multimodal-aware
Best for Experimenting · Model comparison
04

Local LLM

Your machine. Your rules.

Point at Ollama, LM Studio, or any OpenAI-compatible endpoint on localhost. Full control, zero cloud, custom fine-tunes welcome.

  • Ollama / LM Studio
  • OpenAI-compatible
  • Zero cloud dependency
Best for Air-gapped · Custom models

One click to swap.

Your conversation history carries over. Try Gemini Nano for a task, hit a hard one, flip to Claude on OpenRouter, finish, flip back. No reconfiguration.

Auto Browser settings — four LLM provider tabs (Gemini, OpenRouter, Built-in AI, Local) with a backend chooser (Ollama, LM Studio, Custom) and base URL field.
Under the hood

Not magic. Just honest engineering.

Every surface is explainable and auditable. Read the full architecture in the docs, or skim the four things that make it work.

Complete browsing toolkit

Perceive, interact, navigate, observe — everything the agent needs to work a page without guessing. Works on pages built long before AI agents existed.

WebMCP-ready

Pages that describe themselves with WebMCP get preferred treatment. A tagged form beats a dozen DOM traversals — reliable, explicit, fast.

Permission-first

The agent never acts silently. Writes prompt. Sensitive origins are blocked. Redirects invalidate prior approvals. Secrets are redacted before they can leak.

Remembers what it tried

The agent tracks its own steps, notices when a strategy isn't making progress, and summarizes long sessions so it stays coherent turn after turn.

Read the docs
For developers

Your site, agent-ready.

Auto Browser speaks WebMCP — the open web spec that lets your pages describe themselves to any AI agent, not just ours. Tag your forms, expose your actions, and the agent uses your tools instead of poking around the DOM.

  • Open spec, not ours. WebMCP is developed in the open at the Web Machine Learning community group. Auto Browser is one of many agents that will speak it.
  • More reliable than DOM clicks. A tagged form or a registered tool is unambiguous. No selector brittleness, no misclicks, no guessing.
  • Zero dependencies. Auto Browser handles the runtime. You don't ship an SDK or install a package.
Read the WebMCP spec
Open spec

WebMCP

A web-native protocol for pages to expose tools and actions to AI agents. Developed in the open by the Web Machine Learning community group.

github.com/webmachinelearning/webmcp
Privacy & safety

Agent on a leash, not a loose one.

An agent that can click anything is dangerous by default. Auto Browser is designed to be safe first — six layers of protection, starting with the one you control.

  • Every write asks

    Clicks, typing, navigation — the agent requests approval before it acts. Read-only looks (snapshots, reading the page) don't interrupt you; writes always do.

  • Script execution always prompts

    Running arbitrary JavaScript is the most powerful thing the agent can do, so it's also the most gated. Even broad approvals never cover it.

  • Sensitive sites blocked

    Banking, government, and healthcare origins are off-limits by default. The agent is disabled entirely on those sites — there's no way to override inline.

  • Redirect protection

    If a page redirects mid-action (login loop, phishing hop), prior approvals no longer apply. The agent has to ask again for the new origin.

  • Secrets stay secret

    Passwords and tokens are redacted before they can leave your browser. The agent can tell a field has a value, but it never sees the value itself.

  • On-device option

    Gemini Nano runs entirely inside Chrome — no cloud, no account, no network. Pick it and your data never leaves your machine.

app.netlify.com — Project configuration
  1. go to project overview
  2. Check current page for navigation to project overview

  3. take_snapshot

    Done
  4. Click Projects link in sidebar to navigate to project overview

  5. click

    Running Done Denied
  6. Allow Auto Browser to click on app.netlify.com?

  7. Navigation complete. You're on the project overview now.

  8. Navigation complete. I won't prompt again for clicks on this domain this turn.

  9. Action skipped. Tell me what you'd like to do instead.

Try it — choose an option above. You approved this single click. The next write will still ask. Mutating actions on this domain won't prompt again until the turn ends. The agent stops and waits for your next instruction.
Read the full privacy policy
FAQ

Questions, answered.

  • Is Auto Browser free?
    Yes. The extension is free to install. AI provider costs depend on what you pick — Chrome Built-in AI (Gemini Nano) and local LLMs are entirely free; Google Gemini and OpenRouter charge per request with your own API key.
  • Which AI provider should I pick?
    For privacy and offline use, pick Chrome Built-in AI (Gemini Nano). For best quality, pick OpenRouter with Claude Sonnet or Google Gemini directly. For air-gapped work, point it at a local Ollama. You can swap anytime — your conversation carries over.
  • Do I need an API key?
    Only for cloud providers (OpenRouter, Google Gemini). Chrome Built-in AI and local LLMs need no key. Keys are stored locally in your browser — they never reach our servers (because there aren't any).
  • Does my browsing data leave the browser?
    Only what the model needs to answer your turn, and only if you're using a cloud provider. Picking Chrome Built-in AI or a local LLM means zero network traffic. There's no telemetry, ever.
  • Which sites does it work on?
    Any site. The extension uses Chrome DevTools Protocol under the hood, so the agent sees the same rendered page you do — including authenticated content. Banking, gov, and healthcare origins are blocked by default for safety.
  • What are the install requirements?
    Chrome 126+ for basic use. Chrome 138+, macOS 13+ / Windows 10+ / Linux / ChromeOS, 16 GB RAM, and 22 GB free disk are required for the on-device Gemini Nano option.
  • Can I swap providers mid-conversation?
    Yes. Open the settings dropdown, pick a different provider, and keep going. History, attachments, and tool results carry over. There's no re-authentication with the site you're on — Auto Browser is using your existing Chrome session.
  • How do I uninstall?
    chrome://extensions → Auto Browser → Remove. All data (API keys, conversation history, settings) is cleared. No accounts to cancel.
  • How's this different from ChatGPT browsing or Claude for Chrome?
    Those tools route your prompt through their own servers and their own browser. Auto Browser lives in your Chrome — same cookies, same tabs, same login state. It's model-agnostic (you choose), and it respects page-defined WebMCP tools instead of reinventing them.

Stop doing it by hand.

Install free. Pick your favourite AI. Describe the task. Watch it happen.

Install free Read the docs

Free to install. Private by default. No account required.