mcpField
B
browse

browse

github.com/topshark-jimAgent workflows054 installsUpdated 33w ago
Repository

About browse

browse is an agent skill for browser automation with `agent-browser`. It gives agents an explicit command flow and guidance for inspecting and interacting with pages.

Key features

  1. 01Navigate and control browser sessions
  2. 02Wait for pages to finish loading
  3. 03Inspect pages with interactive snapshots
  4. 04Use snapshot references to interact
  5. 05Review console, errors, and network requests

Use cases

  1. 01Automate navigation and interaction on websites
  2. 02Inspect page content when selectors are unclear
  3. 03Troubleshoot pages using browser diagnostics

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/topshark-jim/gstack --skill browse

After installing, type/browsein Claude Code to confirm it loaded.

Details

From the project README

browse

Use agent-browser directly for browser automation. Keep commands explicit and prefer the built-in session model instead of adding a second wrapper layer.

Requirements

  • agent-browser must be available on PATH.
  • If browser binaries are missing, run agent-browser install.
  • Reuse the default session unless isolation matters; close it with agent-browser close when the workflow is finished.

Default Flow

agent-browser open https://example.com
agent-browser wait --load networkidle
agent-browser snapshot -i

Use snapshot -i when selectors are unclear, then interact with @ref targets from the snapshot output.

Command Map

# Open or navigate
agent-browser open https://example.com
agent-browser back
agent-browser forward
agent-browser reload

# Inspect content
agent-browser get title
agent-browser get url
agent-browser get text body
agent-browser get html main
agent-browser snapshot -i

# Interact
agent-browser click @e3
agent-browser fill @e4 "[email protected]"
agent-browser type @e5 "hello"
agent-browser press Enter
agent-browser select @e6 option-a
agent-browser upload @e7 /tmp/file.png
agent-browser wait @e8

# Visual checks
agent-browser screenshot /tmp/page.png
agent-browser screenshot --full /tmp/page-full.png
agent-browser pdf /tmp/page.pdf

# Debugging
agent-browser console
agent-browser errors
agent-browser network requests
agent-browser eval "document.title"

# State and tabs
agent-browser cookies get
agent-browser storage local get
agent-browser tab new
agent-browser tab list
agent-browser tab 2
agent-browser close

Working Rules

  • Prefer wait --load networkidle after navigation-heavy steps.
  • Prefer snapshot -i before click, fill, or select when the page is not stable enough for CSS selectors.
  • Use console, errors, and network requests before concluding that a page is broken.
  • Use --profile <path> when a persistent login or browser profile matters.
  • Use --session <name> when separate browser contexts are required in the same task.
  • If agent-browser is missing, tell the user exactly how to install it before continuing.

Comments

Markdown supported