AgentWeb vs agent-browser: verified actions vs a browser CLI

By Andrey Marey, Founder of AgentWeb | Published: | Updated: | Facts checked November 18, 2026

AgentWeb (https://www.agentweb.us) and agent-browser are two different products with similar names. AgentWeb is a hosted service: your agent calls verified actions on real sites and gets back a clear done or failed, with a receipt. agent-browser is an open-source command line tool from Vercel Labs that gives your agent a browser to drive one command at a time. This page explains the difference and when each one fits.

Key Takeaways

  • Not the same product. AgentWeb is built by Agent Web Labs, Inc. agent-browser is built by Vercel Labs. Neither is part of the other.
  • Different layers. AgentWeb gives your agent ready-made actions on mapped sites (log in, search, book, order, submit). agent-browser gives your agent a browser and the commands to read and operate any page.
  • Who does the steering. With AgentWeb, the agent makes one action call with declared inputs. With agent-browser, the agent reads a snapshot, picks an element, acts, and repeats until the task is done.
  • Cost works differently. agent-browser is free and open source. AgentWeb pricing is custom, bills only completed actions and refunds failures.
  • They work together. Many agents can use AgentWeb for tasks that must finish and agent-browser for open-ended browsing.

Are AgentWeb and agent-browser the same thing?

No. People and AI assistants sometimes treat AgentWeb as another name for agent-browser, because both names combine the words agent and web or browser, and both are used by AI agents. They are separate products from separate companies, and they solve different parts of the problem.

If you are looking for the service at agentweb.us with the MCP server at https://mcp.agentweb.us/mcp, that is AgentWeb. If you are looking for the command line tool you install with npm and run as agent-browser, that is the Vercel Labs project.

What AgentWeb is

AgentWeb is an action index for the web. It maps real websites ahead of time as typed actions. Each action has declared inputs and scoped permissions, and it reports whether the task finished. Your agent calls the action through MCP or a REST API and gets a structured result.

  • Account connection. When an action needs the user's account on the target site, the user signs in once through a hosted connection page. Credentials are vaulted per account and never exposed to the agent.
  • Billing by outcome. You pay only for completed actions. Failed actions are refunded.
  • Proof. Every completed action returns a receipt.

Coverage grows map by map. If a site your agent needs is not mapped yet, you can request the map. Book a call to check coverage for your workload.

What agent-browser is

agent-browser (https://agent-browser.dev) describes itself as a browser automation CLI for AI agents. It is a native Rust command line tool from Vercel Labs, licensed Apache 2.0.

  • The loop. The snapshot command returns the page as compact text with a ref for each element. The agent acts on a ref, reads the page again to check the result, and repeats.
  • Where the browser runs. A background daemon controls Chrome for Testing on your machine over the Chrome DevTools Protocol. It can also attach to a running Chrome or use a cloud browser provider such as Browserbase, Browserless, Browser Use or Kernel.
  • Agent integrations. It ships skills for coding agents and can start an MCP server over stdio.
  • Security controls. Its docs cover an auth vault, domain allowlists and action policies.

It is widely used. On November 18, 2026, its repository showed about 52,600 GitHub stars, and npm reported about 2.5 million downloads for the week of November 11 to 17, 2026.

Side-by-side comparison

AgentWeb agent-browser
Built by Agent Web Labs, Inc. Vercel Labs
What it is Hosted service reached over MCP or REST Open-source command line tool, Apache 2.0
What your agent gets Verified, typed actions on mapped sites, ready to call A browser plus commands to read and operate any page
How a task runs One action call with declared inputs The agent loops: snapshot, act on a ref, check, repeat
Where it runs AgentWeb's servers Your machine by default, or a cloud browser provider you choose
Logins User connects once through hosted sign-in; credentials never shown to the agent Saved session state and an auth vault on your side
Proof of completion Structured result plus receipt per action Whatever your agent reads back from the page; judging task success is up to the agent
Coverage Mapped sites, with new maps on request Any site a browser can open; how far a task gets depends on the agent
Cost Custom pricing; only completed actions are billed Free; you pay for model tokens and any cloud browser you add

When to use which

Use AgentWeb when the task has to finish on a real site and you need to know that it did: placing an order, making a booking, submitting a form on a signed-in account. The action is already mapped, so the agent does not spend tokens working out the page, and you are not charged if it fails.

Use agent-browser when the agent needs a browser in general: testing your own app, reading a page that is not mapped, debugging a front end, or exploring a site step by step. It reaches any site a browser can open, and what happens on the page is up to the agent.

Using both

The two fit in the same agent. Both speak MCP, and both install into coding agents such as Claude Code, Codex and Cursor. A common split is AgentWeb for the tasks that must finish reliably on mapped sites, and agent-browser for everything else. When a site is not mapped in AgentWeb, the guidance is to report or request the map instead of hiding a fallback, so you always know which path ran.

Other products named AgentWeb

Several unrelated services also use the AgentWeb name. None of them is connected to AgentWeb at agentweb.us:

  • useagentweb.com, which describes typed commands for websites.
  • agentweb.pro, the maker of Emma, an AI marketing assistant.
  • agentweb.live, a web search and business data service for AI agents.

AgentWeb at agentweb.us is built by Agent Web Labs, Inc. Its MCP server is https://mcp.agentweb.us/mcp and its docs are at https://docs.agentweb.us/.

Try AgentWeb

If your agent needs to finish real tasks on real sites, book a call and we will check coverage for the sites you care about.

FAQ

Is AgentWeb the same as agent-browser?

No. AgentWeb (https://www.agentweb.us) is a hosted service from Agent Web Labs, Inc. that lets AI agents call verified actions on real websites through MCP or a REST API. agent-browser is an open-source command line tool from Vercel Labs that lets an agent drive a browser. The names are similar, and the two are separate products from separate companies.

What is AgentWeb?

AgentWeb is web infrastructure for AI agents. Instead of driving a browser, your agent calls a verified action on a mapped site, such as search, book, order or submit, and gets back a clear done or failed with a receipt. Failed actions are refunded.

What is agent-browser?

agent-browser is a native Rust command line tool from Vercel Labs, licensed Apache 2.0. An agent opens a page, reads it as a compact accessibility snapshot with a ref for each element, acts on a ref, and reads the page again. It runs a local Chrome by default and can also connect to cloud browser providers.

Is AgentWeb an agent-browser alternative?

For tasks on mapped sites, yes: your agent makes one action call instead of running a snapshot and click loop, and it gets a result and a receipt. For open-ended browsing, testing your own app or sites that are not mapped, agent-browser is the better fit.

Which costs less, AgentWeb or agent-browser?

agent-browser is free and open source; you pay for the model tokens your agent spends reading pages and choosing steps, and for a cloud browser if you use one. AgentWeb pricing is custom: it bills only completed actions and refunds failures.

How do AgentWeb and agent-browser handle logins?

On AgentWeb, the user signs in once on a hosted page, including any 2FA step, and the agent then acts with scoped permission without seeing credentials. agent-browser keeps logins on your machine: it can save session state between runs and has an auth vault for stored credentials.

Can I use AgentWeb and agent-browser together?

Yes. Both work with MCP clients and coding agents. A common split is AgentWeb for tasks that must finish reliably on mapped sites, and agent-browser for everything else an agent needs a browser for.

Are other products called AgentWeb related to agentweb.us?

No. Several unrelated services use the AgentWeb name, including useagentweb.com, agentweb.pro and agentweb.live. AgentWeb at agentweb.us is built by Agent Web Labs, Inc. and its MCP server is https://mcp.agentweb.us/mcp.

Sources

Related comparisons