Keep the agent you run. Take the keys off it.

AgentValet sits between your agent and the platforms it calls. Your agent keeps its model and framework; only how it connects differs. Find your runtime below and follow its page.

Two ways to connect

Both end in the same place: an agent with its own identity, nothing granted until you grant it, and every call on the record. The difference is where the identity lives.

The CLI: a key on your machine

One command signs you in through your browser, creates the agent, keeps its private key on this machine, and writes the MCP config for every host it finds. You don’t copy any ids.

Terminal
npx @agentvalet/register
  • Used by Claude Code, Cursor, Codex, Factory Droid and governed git.
  • The key lands in .agentvalet/agent.key in the directory you ran it from. Add --project to write project config instead of user-wide config.

OAuth: approve in the browser

Add one URL to the app, approve in your browser, and AgentValet creates a new agent for that connection. There’s no key file to keep, and the app handles the token.

Your MCP server URL
https://api.agentvalet.ai/mcp/v1/<owner-id>
  • Your owner id is in the dashboard under Settings, Account. The Connect Claude, ChatGPT, Copilot… button on the Agents page builds the URL for you.
  • Used by Claude Desktop and Hermes, and works for Claude Code, Cursor and Codex too. The same panel has steps for Claude on the web and ChatGPT.

Not on the list?

Most agents can still go through AgentValet today.

Your agent keeps its job. It loses the keys.

Register one agent, grant it only what it needs, and approve the risky calls from your phone. Free to start, no credit card.