chrome-use
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
No arguments | |||
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
Use the chrome_use_* tools to drive a real browser. Call chrome_use_open to navigate, then chrome_use_snapshot to get an accessibility tree with @ref handles before clicking/typing. This is the CORE tool profile (~13 tools) — not full chrome-use CLI parity. Start the server with --tools all for the extended profile.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| chrome_use_openA | Navigate to a URL. Omitting |
| chrome_use_readB | Fetch a URL (or the active tab, if url is omitted) as agent-readable text/markdown. |
| chrome_use_snapshotA | Accessibility tree of the current page with @ref handles, for locating elements to click/fill/type. |
| chrome_use_clickB | Click an element (by CSS selector or @ref) or a raw viewport coordinate. Provide either |
| chrome_use_fillA | Clear and fill a form field (handles CodeMirror/Monaco/ProseMirror/contenteditable in addition to plain inputs). |
| chrome_use_typeC | Type text into an element (or, with no selector, whatever currently has focus). |
| chrome_use_pressC | Press a keyboard key (e.g. Enter, Tab, Control+a). |
| chrome_use_eval | Run JavaScript in the page's main world (state persists, page handlers fire) and return its result. Over the extension relay one evaluation, including an awaited promise, has about 8 s regardless of timeoutMs; a longer script keeps running in the page but its result is lost, so split long waits into separate calls. |
| chrome_use_waitA | Wait for an element, a fixed duration, a URL, a load state, a JS condition, text, or a download. Provide exactly one of: selector, ms, url, loadState, expression, text, download. |
| chrome_use_backB | Go back in browser history. |
| chrome_use_forwardB | Go forward in browser history. |
| chrome_use_reload | Reload the current page, or the tab named by tabId. |
| chrome_use_confirmA | Answer an action the CLI held with confirmation_required (a sensitive site, a payment card field, a purchase, or a rule the user or their organization set, such as downloads or password fields). Tell the user what you were about to do and where first. With approval: user, only the user can allow it in the chrome-use extension; this call then waits up to 40s for their choice and runs the action if they allowed it. decline: true withdraws the request. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 13 tools
Most tools have clearly distinct purposes: navigation, history, inspection, interaction, waiting, and confirmation are separable. The main potential confusion is between fill and type, and between read and snapshot, but the descriptions clarify that fill clears while type does not, and read returns text while snapshot returns an accessibility tree.
All tool names follow a consistent chrome_use_verb pattern with lowercase snake_case verbs. The convention is predictable and readable throughout.
With 13 tools, the set is well-scoped for browser automation: core navigation, history, interaction, waiting, inspection, and confirmation are covered without excessive fragmentation. Each tool earns its place.
The surface covers the main browser-use lifecycle: open, read, snapshot, click, fill, type, press, wait, eval, back/forward/reload, and confirmation handling. Minor gaps include explicit hover, select-option, drag, screenshot, and tab/window management, but most can be worked around with existing tools such as eval or coordinate clicks.