MCP tools: remote-agent-ops-hosting
remote-agent-ops-hosting
Registered when: platform: mcp.agentOps + managed hosting covering the key's org (agents.managedHosting and agents.launch, both all, on by default); key names agent_launch, org admin, never a runner key.
agent_hosting_kill
Trip the org kill switch for managed agent hosting: the org's host cap is set to 0 and every live managed host is destroyed (its agents stop). Making things safer is allowed to agents; turning hosting back on is a person's job. Irreversible for the running work, so it needs confirm: true (or dryRun: true to see what would stop). Same key rules as agent_hosting_status.
| Parameter | Type | Required | Description |
|---|---|---|---|
orgId | string | yes | Organization id (uuid). |
confirm | boolean | no | Must be true to actually trip the switch. |
dryRun | boolean | no | true = check and plan only: makes the read calls that prove the target exists and is visible to this key, returns the writes it WOULD make, writes nothing. Write permission is only decided when the write runs. |
Example call:
{"name":"agent_hosting_kill","arguments":{"orgId":"<orgId>"}}
Related: agent_hosting_status.
agent_hosting_status
Read an org's managed agent hosting (where QA Commander and other hosted agents run): each host's state, the host and host-hour caps with this month's usage (a host cap of 0 means the kill switch was tripped or hosting is off). Read-only. Needs a key naming the 'agent_launch' family and an org admin; runner keys are refused.
| Parameter | Type | Required | Description |
|---|---|---|---|
orgId | string | yes | Organization id (uuid). |
Example call:
{"name":"agent_hosting_status","arguments":{"orgId":"<orgId>"}}
Related: agent_hosting_kill.