mcpField
A
agent-browser-wsl2-setup

agent-browser-wsl2-setup

github.com/arthurzakirovAgent workflows1162 installsUpdated 2w ago
Repository

About agent-browser-wsl2-setup

A setup skill for running agent-browser in WSL2, covering installation and runtime configuration. It distinguishes native Linux Chrome from a Windows host browser bridge and provides separate steps for each path.

Key features

  1. 01Install agent-browser and its runtime
  2. 02Configure the runtime setup script
  3. 03Test native WSL2 Chrome browsing
  4. 04Connect to a Windows browser bridge
  5. 05Troubleshooting and architecture references

Use cases

  1. 01Preparing agent-browser on a new WSL2 machine
  2. 02Testing browsing with native Linux Chrome
  3. 03Attaching agent-browser to a Windows browser
  4. 04Diagnosing agent-browser setup failures

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/arthurzakirov/agentdesk --skill agent-browser-wsl2-setup

After installing, type/agent-browser-wsl2-setupin Claude Code to confirm it loaded.

Bundled files

References and scripts installed along with SKILL.md.

agents/
openai.yaml
references/
architecture.md
dead-ends.md
troubleshooting.md
scripts/
connect-windows-browser.sh
ensure-wslconfig.ps1
setup-agent-browser-runtime.sh
setup-agent-browser-wsl2.sh
setup-native-wsl-chrome.sh

Details

From the project README

Agent Browser WSL2 Setup

This skill owns agent-browser setup only.

Browser availability belongs in $wsl2-browser-setup. Use that skill first if neither of these is already true:

  • native Linux google-chrome inside WSL2 works for real browsing and sign-in
  • a Windows host browser bridge is reachable from WSL2 on /json/version

Bootstrap Split

On a new machine, run these two commands directly from the AI:

npm install -g agent-browser
agent-browser install

Then have the human run this interactive command:

npx skills add vercel-labs/agent-browser

Important:

  • npm install -g agent-browser is non-interactive and should be run by the AI when possible
  • agent-browser install is non-interactive and should be run by the AI when possible
  • npx skills add vercel-labs/agent-browser is interactive and should be run by the human

After the human finishes the interactive npx skills add ... flow, continue with the runtime setup script.

Runtime Setup

Run:

bash "${CODEX_HOME:-$HOME/.codex}/skills/agent-browser-wsl2-setup/scripts/setup-agent-browser-runtime.sh"

The script:

  • installs agent-browser if missing
  • runs agent-browser install if needed
  • clones vercel-labs/agent-browser if missing

Mode A: Native WSL Browser

If the browser path from $wsl2-browser-setup is native Linux Chrome inside WSL2, test agent-browser directly:

agent-browser doctor --offline --quick
agent-browser open https://example.com
agent-browser snapshot -i -c

If this still hangs even though manual browsing works, stop debugging the browser here and use Mode B.

Mode B: Windows Host Browser Bridge

If the browser path from $wsl2-browser-setup is a Windows host browser bridge, attach agent-browser to that existing CDP endpoint:

bash "${CODEX_HOME:-$HOME/.codex}/skills/agent-browser-wsl2-setup/scripts/connect-windows-browser.sh" --session windows-host --bridge-port 9333
agent-browser --session windows-host get title
agent-browser --session windows-host snapshot -i -c

This was the proven control path before the native browser path was cleaned up.

Legacy Combined Wrapper

scripts/setup-agent-browser-wsl2.sh still exists as a compatibility wrapper for the older one-shot Windows bridge flow.

Prefer the split skills instead:

  1. $wsl2-browser-setup
  2. $agent-browser-wsl2-setup

Bundled Resources

Comments

Markdown supported