Overview
Melian can operate a web browser autonomously, opening pages, clicking elements, filling forms, extracting content, and taking screenshots. This enables web-based research, form submission, and visual verification of page state. The agent-browser binary is spawned as a subprocess per tool call via Bun.spawn(). Each tool call has a 30-second default timeout.
Tools
| Tool |
Parameters |
Description |
browser_open |
url: string |
Navigate to a URL and return an accessibility tree snapshot |
browser_snapshot |
- |
Return an accessibility tree snapshot of the current page state |
browser_click |
ref: string |
Click an element by its @ref identifier from the accessibility tree |
browser_fill |
ref: string, text: string |
Type text into a form field identified by its @ref identifier |
browser_extract_text |
selector?: string |
Extract visible text from the page or a specific element |
browser_screenshot |
- |
Capture a PNG screenshot of the current viewport |
browser_close |
- |
Close the browser session and release the binary |
Tool Parameter Details
browser_click
| Parameter |
Type |
Required |
Notes |
ref |
string |
yes |
A @ref identifier from the accessibility tree snapshot |
browser_fill
| Parameter |
Type |
Required |
Notes |
ref |
string |
yes |
A @ref identifier for the input field from the accessibility tree |
text |
string |
yes |
Text to type into the field |
browser_extract_text
| Parameter |
Type |
Required |
Notes |
selector |
string |
no |
Omit to extract full page text; provide to scope to a subtree |
Architecture
The agent-browser binary is spawned as a subprocess per tool call via Bun.spawn(). The 30-second timeout applies per tool call; long-running navigations that exceed this will return an error.
Browser automation is used by the self-healing pipeline for visual verification of deployed changes, and is available as a general-purpose tool in conversations for any web-based task.