Repository navigation
Conversation
Browser Use Pi behind the A2A agent framework: each turn runs agent.mjs once, which drives a Browser Use Cloud browser (or a cdp_url) with the env's MCP tools beside it and calls the [model] endpoint as an OpenAI-compatible one; a context's turns continue one conversation. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Review the following changes in direct dependencies. Learn more about Socket for GitHub.
|
|
Warning Review the following alerts detected in dependencies. According to your organization's Security Policy, it is recommended to resolve "Warn" alerts. Learn more about Socket for GitHub.
|
Adds Browser Use as a shipped agent, as agreed with @EdgarArakelyan in Slack.
Each turn runs
agent.mjsonce with@browser_use/pi. It drives a Browser Use Cloud browser, or any Chrome DevTools endpoint passed ascdp_url, and it gets the task env's MCP tools as<server>__<tool>. It calls the[model]endpoint as an OpenAI-compatible one, withLITELLM_EXTRA_HEADERSincluded, and a context's turns continue one conversation through a saved history file. It's two files,a2a_server.py(151 lines) andagent.mjs(77 lines), plus registration likeletta-code.Validator (
agent-env a2a-agent put, local Docker,gpt-5.4-mini):message/send,tasks/get, agent-config identity, role, system prompt, text input, MCP tool calls onmail, live trajectory, tool-call countinstall/v1, non-text inputUnit tests in
tst/unit/agents/browser_usecover one turn, an early stop, a crash and a timeout kill (24 passed with the bundle and image tests). A cancel or timeout sends SIGTERM first, so the agent stops its cloud browser before it is killed; this was checked live (browser went fromactivetostopped). To use it, the registration needsBROWSER_USE_API_KEY.