How it connects
The address is your hub plus /mcp. Settings shows it, with the line to add it to Claude Code. Any MCP client that speaks streamable HTTP works.
It signs in with a user's API key (Authorization: Bearer). The agent can do exactly what that user can, in that user's companies, and nothing more. We advise a dedicated user for each agent. Every tool call is recorded in the audit trail under that user.
It can't open the screen, files or a terminal. Those stay with a signed-in person.
What approval means here
Changes to machines (a BE AI proposal, a group security fix, a saved script) run only when approved. The approval is recorded under the key's user. If that user is allowed to approve, an agent using their key can approve too. Give an agent a user without approval rights if you want a person to approve every change.
The 38 tools
- Machines: who am I, list machines, machine detail, machine history, fleet summary, events, patch status.
- Alerts: list, acknowledge, resolve, comment, open a ticket, change a machine's alert settings and reboot windows.
- BE AI: ask BE AI, list runs, run detail, list proposals, approve or reject a proposal.
- Scripts: list saved scripts, run a saved script.
- Tickets: list, detail, raise, reply or add an internal note, update, log time, assignees, categories and canned replies, accept BE AI triage, investigate, add to the knowledge base, mark spam.
- Knowledge base: search, write.
- Security: overview, findings (filter by company, machine, rule or severity), one machine's security, accept, reopen, assign or mark done a finding, check a machine now, leave a machine out or put it back, propose a group fix, approve or reject a group fix, BE AI action plan.
- Reports: list, and run any report.
- What's pending: the digest.