MCP admin tools
When a platform administrator connects an AI assistant through MCP, it gets everything a Pro user's assistant has plus a set of administrator tools that work across other people's accounts. By the end of this page you will know what those tools are, what to ask for, how they are recorded, and the actions that are deliberately left to the admin panel.
How users connect a client, and what the ordinary tools do, is covered in AI & MCP. Connecting as an admin works the same way — see Connect an AI assistant.
How admin tools differ
- They are named apart. Every cross-account tool starts with
admin_. The ordinary tools still act on your own account:list_labsreturns your labs,admin_list_labsreturns everyone's. - They appear only for administrators. A connection is given admin tools when the signed-in account is a platform admin. Nobody else sees them.
- They are recorded like everything else. Each call lands in your
history — the MCP page's Activity tab and the assistant's
my_history. Your history shows your account only; there is no tool that reads another person's MCP activity, for admins either. - The scope rule still applies. Requests outside labs and their resources are refused for admins exactly as they are for users.
Admin connections also see labs still in alpha, and four template tools
users do not have: transfer_template_owner, set_template_free_plan,
publish_template and unpublish_template (listing a template on the
LabStore). See Share and publish.
An assistant with admin tools can change organizations you do not belong to and approve work that affects other users. Connect admin clients deliberately, approve only clients you started, and disconnect them when you are done.
Users and labs
What you can ask
- "Find the user ravi and show me their plan and labs."
- "List every running lab on the platform."
- "Which users are using the most storage?"
- "run_command keeps failing for several people — is the platform able to run commands at all?"
| Tool | What it does |
|---|---|
admin_find_user | Finds an account by username or email |
admin_list_labs | Labs across the platform, not just yours |
admin_storage_usage | Storage use across users |
admin_exec_health | Checks whether running commands inside labs works platform-wide. It runs nothing in anyone's lab |
admin_exec_health is the one to reach for when commands fail the same way
for more than one lab or person — it tells a platform problem apart from a
lab problem in one call.
Account-level changes — plans, limits, blocking, impersonation — stay in the admin panel. See Users.
Organizations and departments
What you can ask
- "List all organizations and their owners."
- "Create an organization called Acme Labs owned by priya@example.com."
- "Add a department called Data Science under acme-labs."
- "Make Ravi the owner of acme-labs."
- "What would tearing down acme-labs remove?"
| Tool | What it does |
|---|---|
admin_list_entities | Every organization and department |
admin_entity_get | One organization's or department's details |
admin_entity_storage | Its storage use |
admin_create_org | Create an organization |
admin_create_department | Create a department under one |
admin_update_entity | Change its details or limits |
admin_set_entity_owner | Change its owner |
admin_add_entity_member / admin_remove_entity_member | Add or remove a member in any organization |
admin_teardown_preview | Lists everything a teardown would destroy — labs, devices, addresses, domains, service users, grants, storage and child departments. Destroys nothing |
An assistant can preview a teardown but cannot run one. Tearing down an organization is done by hand in Admin Panel → Organizations → Manage → Danger → Teardown…, with the typed confirmation. Do not archive an organization as a step towards deleting it. See Tearing down an organization.
Who owns what, and platform totals
What you can ask
- "Who owns shop.selfmade.one?"
- "Which account has the address 203.0.113.24?"
- "How many domains, addresses and devices are in use across the platform?"
| Tool | What it does |
|---|---|
admin_who_owns_domain | The account or organization holding a domain |
admin_who_owns_address | The owner of a private network address |
admin_domain_stats | Domain counts across the platform |
admin_address_stats | Address allocation across the platform |
admin_device_stats | VPN device counts across the platform |
Storage
What you can ask
- "Show me the ten heaviest storage users."
- "How much storage does acme-labs use against its limit?"
| Tool | What it does |
|---|---|
admin_storage_usage | Usage across users |
admin_entity_storage | Usage for one organization or department |
Changing limits and moving accounts between storage pools is done on Storage quotas.
Certificate re-issue review
Users can ask their assistant to request a replacement for a broken certificate. Those requests queue for an administrator. Your assistant can help you review the queue.
What you can ask
- "Show me the certificates that are expired or failing to renew."
- "What re-issue requests are pending? Explain each one."
- "Approve the pending requests for the shop and blog domains; reject the rest."
- "Queue the stale leftover certificates for cleanup — but show me the list first."
- "Is there an approved batch waiting to be applied?"
| Tool | What it does |
|---|---|
admin_ssl_list | Every certificate on the platform, with status and expiry |
admin_ssl_cert_details | One certificate's domains and history |
admin_ssl_requests | Re-issue requests, filtered by state; can also show what would be queued as leftover cleanup |
admin_ssl_decide | Approve or reject pending requests. Reports a result for each request |
admin_ssl_request_delete | Remove a request from the queue |
admin_ssl_queue_residue | Queues every certificate that is failing only because of names no lab uses any more. They land already approved — review the list first |
admin_ssl_apply_status | Whether an approved batch is waiting, and how the last one went |
Good to know
- Approving does not replace anything. It moves the request to Approved, where it waits for you.
- Approval re-checks the request. If the certificate has recovered, or a domain no longer points at the platform, the request is closed as stale. That is the right outcome, not an error.
- Batches are not all-or-nothing. Read the per-request results; three of five succeeding is still a successful call.
Applying is a dashboard action
There is deliberately no tool that applies an approved batch, for anyone. When the queue is ready:
- Open Admin Panel → Platform → SSL Queue.
- Go to the Approved tab and check the list.
- Click Apply Approved.
Ask your assistant to prepare and approve the batch, then apply it yourself. See Certificates and VPN.

Next
- AI & MCP — what every connected assistant can do
- Certificates and VPN — the SSL Queue in the admin panel
- Organizations — managing organizations by hand