The agent directs the work. Configure the proxy in the tool making the web request.
Set the connection in the tool your agent uses
The agent plans the task, its tools run it, and the configured proxy provides the network connection. A management API or MCP server can supply tools without carrying the task’s web traffic.
Start with read-only work. Verify the account and proxy resources, then configure the browser or script. Limit the first run to a few requests and check the returned pages before scheduling a larger job.
Follow one web request
Start with the action you want to perform: opening a public page, fetching HTML or checking a local storefront. Identify whether the request comes from a browser, a script or a hosted tool.
Use that tool’s documented proxy options. A proxy setting in your chat application may not apply to a separate browser or remote tool. Some tools do not expose custom proxy settings.
Ask for a result you can verify
When testing product research, save the displayed product name, price, currency, source URL and the time you checked it. Check the extracted values against the original pages before collecting more.
Mark access errors and missing values separately from valid results. Use a sample you can check by hand, then plan the next run using the traffic consumed and the proportion of tasks completed.
Check proxy, model and browser charges separately
Proxy traffic, model calls and hosted browser usage can have different charges. A single prompt may load several pages and retry requests; prompt count alone cannot estimate GB consumption.
USM’s public API, installable Agent Skills and MCP service are not released. The proxy setup guides do not include those packages.