How to verify browser agent results
What each fastbrowse run status and exit code means, how citations back an answer, and what to do when a run ends unverified.
blog
Browser agent guides and notes
How fastbrowse runs, reports and stays within limits. For flags and setup, read the docs.
Results and cost
Cap browser automation cost per run
What a fastbrowse run reports it spent, what each charge is for, and how to cap a run with --max-dollars and --max-steps.
How the agent acts
Why an AI agent clicks the wrong button
An agent that clicks screen coordinates can hit a control that is not there. How an indexed control set bounds what fastbrowse can click.
Browsers and logins
Connect an agent to a running Chrome over CDP
Drive a Chrome or Electron app you already have open over its DevTools port: the launch flag, picking a window, and keeping a signed-in browser open.
Browser agent logins without sharing passwords
Three ways to let a fastbrowse run act in your account: a signed-in browser profile, a cloud profile, or a login typed from a Bitwarden vault.
Coding agents and MCP
Add a browser skill to Claude Code and Codex
Install the fastbrowse skill so Claude Code or Codex can browse a site and come back with an answer, its quotes and a run status.
MCP browser server with spending limits
fastbrowse-mcp serves one browse tool over MCP with its limits set on the server. What each flag fixes, and what a calling model can and cannot change.