Summary
The browsing happens elsewhere and your side holds nothing but a task ID.
Submission and collection being separate calls is what keeps a long task from occupying the conversation while it runs. The task goes in as written — the tool is explicit that continuous, complex work does not need to be decomposed before it is handed over. Which model runs it is a property of the task rather than a server-wide setting, so two tasks in the same session can use different ones.
What it is
A two-call interface to 302AI's browser agent: submit a task, then fetch its result.
What you get
- A browser task created from a plain description, on a model you name, without splitting the work into steps first.
- A task ID returned immediately, and the result fetched against that ID once the run has produced one.
Requirements
A 302AI account and its key, supplied as 302AI_API_KEY.
Setup effort
One command plus a key — npx -y @302ai/browser-use-mcp, then supply credentials