Agent Browser AI MCP integration
Give agents goal-level browser instructions instead of step-by-step clicks. Agent Browser AI can act on a task, extract structured page data, and find relevant page elements while preserving governed sessions, human handoff, and user-owned model credentials.
Sign in, add Agent Browser AI in Apps, then choose the account your AI can use.
Tools
| Name | Description |
|---|---|
snapshotCapture the page accessibility tree with element refs. Read this, then act by ref. | Capture the page accessibility tree with element refs. Read this, then act by ref. |
read_textGet the visible text of the page or a target element. | Get the visible text of the page or a target element. |
read_htmlGet the HTML of the page or a target element. | Get the HTML of the page or a target element. |
screenshotTake a screenshot of the page or an element. Returns inline image content, with an optional Filesystem URL. | Take a screenshot of the page or an element. Returns inline image content, with an optional Filesystem URL. |
run_stepsRun several deterministic browser operations in one session round trip. Use for scripted navigation and extraction flows. | Run several deterministic browser operations in one session round trip. Use for scripted navigation and extraction flows. |
current_pageGet the current URL and title. | Get the current URL and title. |
navigateGo to a URL. | Go to a URL. |
navigate_backGo back to the previous page. | Go back to the previous page. |
clickClick an element by ref from snapshot or selector. | Click an element by ref from snapshot or selector. |
typeType text into an element by ref or selector. Optionally submit with Enter. | Type text into an element by ref or selector. Optionally submit with Enter. |
fill_formFill multiple fields at once. Each field has a target, type, and value. | Fill multiple fields at once. Each field has a target, type, and value. |
select_optionSelect one or more options in a dropdown. | Select one or more options in a dropdown. |
hoverHover over an element. | Hover over an element. |
dragDrag one element onto another. | Drag one element onto another. |
press_keyPress a keyboard key, such as Enter, Backspace, ArrowDown, or a. | Press a keyboard key, such as Enter, Backspace, ArrowDown, or a. |
file_uploadSet a file on an upload input. | Set a file on an upload input. |
evaluateRun a JavaScript function on the page or a target element and return its result. | Run a JavaScript function on the page or a target element and return its result. |
wait_forWait for text to appear/disappear with an optional timeout in seconds. With only time, sleep for that many seconds. | Wait for text to appear/disappear with an optional timeout in seconds. With only time, sleep for that many seconds. |
handle_dialogAccept or dismiss a pending JavaScript dialog. | Accept or dismiss a pending JavaScript dialog. |
tabsList, create, close, or select browser tabs. | List, create, close, or select browser tabs. |
show_live_viewMint a watch-only live view URL for the current hosted browser session without handing control away from the agent. | Mint a watch-only live view URL for the current hosted browser session without handing control away from the agent. |
request_humanHand control to a human for login, MFA, or CAPTCHA. Hosted sessions return a viewer URL; local sessions use the local browser window. | Hand control to a human for login, MFA, or CAPTCHA. Hosted sessions return a viewer URL; local sessions use the local browser window. |
resumeTake control back from the human and continue automating. | Take control back from the human and continue automating. |
ensure_connectedPreflight browser readiness. For local browser sessions, briefly waits for the extension or companion and returns whether to proceed or recover. | Preflight browser readiness. For local browser sessions, briefly waits for the extension or companion and returns whether to proceed or recover. |
session_statusGet the current handoff state and page. | Get the current handoff state and page. |
start_sessionStart a fresh browser session. Hosted sessions return a viewer link; Chrome extension sessions wait for the extension; managed-profile sessions return a companion command. | Start a fresh browser session. Hosted sessions return a viewer link; Chrome extension sessions wait for the extension; managed-profile sessions return a companion command. |
end_sessionEnd the browser session and snapshot its profile. | End the browser session and snapshot its profile. |
actUse the configured model provider to complete a goal-level browser instruction. | Use the configured model provider to complete a goal-level browser instruction. |
extractUse the configured model provider to extract structured data from the current page. | Use the configured model provider to extract structured data from the current page. |
observeUse the configured model provider to identify page elements relevant to an instruction without acting. | Use the configured model provider to identify page elements relevant to an instruction without acting. |
Quick Start
import { WeavzClient } from '@weavz-io/sdk'
const weavz = new WeavzClient({ apiKey: 'wvz_...' })
const result = await weavz.actions.execute('agent-browser-ai', 'snapshot', {
workspaceId: '550e8400-e29b-41d4-a716-446655440000',
integrationAlias: 'agent-browser-ai',
input: { /* ... */ },
})Related Integrations
Works well with
Frequently Asked Questions
Is there a Agent Browser AI MCP server through Weavz?
Yes. Add the Agent Browser AI integration to a Weavz workspace, enable the actions you need and connect any required account. Weavz provides the workspace MCP endpoint in Tool Mode or Code Mode. The catalog currently lists 30 Agent Browser AI actions.
What can an AI agent do with Agent Browser AI?
Available actions include Snapshot Page, Read Text, Read HTML, Screenshot, Run Steps, and more. Your workspace exposes only the actions you enable. The agent also needs the intended account access and any required input values.
How do I connect Agent Browser AI to Claude or ChatGPT?
Use a compatible remote MCP client, add the Weavz connector and choose a workspace. Add Agent Browser AI, connect the required account and review the enabled actions. Client availability and account requirements vary; adding the connector does not authorize every app automatically.
Can I use Agent Browser AI through an API or SDK?
Yes, supported actions can run through the Weavz REST API, TypeScript SDK or Python SDK. Configure a workspace and the required connections, use scoped access, and pass the intended end-user context when acting for a customer.
What authentication does Agent Browser AI require?
The catalog lists custom credentials for Agent Browser AI. Use Hosted Connect to authorize the intended account. Required permissions, provider app access and available connection methods depend on the integration.
How do I review sensitive Agent Browser AI actions?
Limit enabled actions, scope credentials to the intended user and configure Human Gates for the operations that need review. A matching policy can pause a send, record change or other sensitive action for an approver before execution.
What limits or costs apply to Agent Browser AI?
Weavz plan limits apply to the actions and resources you use. The external provider may also require an account, a paid plan, API credits or specific permissions, and may enforce its own rate limits. Check current Weavz pricing and the provider’s terms before rollout.
Connect Agent Browser AI in minutes
Get 20,000 free action executions every month. No credit card required.