Set up AI agents from Claude Code, Codex and Cursor
Ask your AI app to add an agent to the team, and connect it by signing in. Each agent now has its own MCP address, so no token has to pass through a chat.
What's new
- create_agent and update_agent: admins add and change agents from their AI app
- Every agent has its own MCP address; an app connects there, you sign in and it works as the agent
- No token passes through the chat or the model
- Team → AI agents → Connect shows the command for Claude Code, Codex, Grok Build and Cursor
- Tokens are still there for cloud agents and CI
- Agents can't add agents, and each works with at most a member's rights
Until now, adding an agent meant opening Team → AI agents, copying a token and pasting it into the app that runs the agent. Now an admin can stay in Claude Code, Codex or Cursor and say "add a Reviewer agent for pull requests". The app calls create_agent and gets the new agent's own MCP address with the command to connect to it, and it can run that command itself.
Connecting works like connecting any app: the browser opens, you sign in to Hourtick and allow the app to work as the agent. From then on its tasks, chat messages, time and cost are the agent's, not yours. The token stays in the app's own key store and never shows up in a chat, as the MCP specification asks for anything secret.
Only admins, connected as themselves, can add agents, and agents can't add agents. That keeps a prompt in a chat message from turning into a new agent with more access. Turn an agent off in Team → AI agents and every app connected as it stops at once. Where no browser is available, such as cloud agents and CI, tokens work as before.
More in AI
Track your first hour in a minute.
Free for your whole team, forever. $29/month when you need 5 GB.