Set up Claros with your agent.
One command, where your agents run. It signs you in through the browser (new here? make your account there) and enrolls this machine's agents.
One command
npm · getclaros$ npx getclarosGuards Claude Code, Codex and Cursor: native hooks check shell commands and file edits, and a local proxy in front of each app's MCP server checks its write calls (reads pass through). It also adds the Claros MCP server. A stopped call never reaches the app; the agent gets the rule, the owner's words and who to ask. Every config is backed up first; npx getclaros undo restores them.
- check it
- npx getclaros testShows a stop.
npx getclaros statuslists what's guarded. - browser agents
- await guard(page)Run the command in a Playwright or browser-use project and it installs the hook. Docs
Or let your agent do it: paste this into Claude Code, Codex or Cursor.
prompt for your agent
Set up Claros for me. 1. Read https://claros.sh/llms.txt, then follow https://claros.sh/agents.md. 2. Run `npx getclaros` in a terminal I can see. It opens my browser so I can sign in (or make an account), then guards the agents on this machine and adds the Claros MCP server. Never ask me for a key. 3. Tell me what it changed, and which apps to restart. `npx getclaros test` shows a stop; `npx getclaros undo` puts every config back. 4. After the restart, list the Claros tools you have (check_action, get_decision, …) and tell me. 5. From now on, call check_action before you save, submit, send or change anything in a business app. - result "stop": don't do it. Tell me the rule, whose rule it is and their words. - result "ask": don't do it yet. Claros has asked the rule's owner. Call get_decision with the check_id (wait 30), again while it's open, and go ahead only if proceed is true, with the action unchanged. Tell me who you're waiting for. - result "nudge": go ahead and apply its suggested fix.