AI Administration (MCP)
Gigantics includes an MCP server. MCP (Model Context Protocol) is the standard way for AI assistants such as Claude, Cursor or VS Code to use external tools. Once your assistant is connected, you can administer Gigantics by asking for what you need in plain language:
- "Check the health of the instance."
- "Why did last night's jobs fail?"
- "Add a PostgreSQL datasource to the Sales project, test the connection and scan it."
- "Create a user for ana@example.com and give her access to the Sales project."
- "Who changed the authentication settings this week?"
The assistant does the work with the same operations that the Gigantics interface and the command line use. It cannot do anything that the account it acts as is not allowed to do, and it cannot make destructive changes without a person approving them.
AI Administration is for administrators. To let an application read a dataset or trigger a pipeline, use a project API key instead.
What the assistant can do
The assistant has more than 100 tools, grouped by what they manage:
| Area | Examples of what you can ask |
|---|---|
| Instance | Check health and version, see which sign-in methods are enabled, read and change the Global Settings, test an LDAP connection |
| Licensing | List licenses, assign or free a license on a tap, cancel or remove a license |
| Organizations and projects | Create, edit, clone, move and delete organizations and projects |
| Users and access | Create users, make a user Root, reset a password, add members to a project, send invitations, create custom roles |
| Datasources and destinations | Add, edit, test and delete taps and sinks, run a test query, sample data |
| Models and datasets | List models, read a schema or its DDL, browse and rename datasets |
| Rules and pipelines | Write, validate and save rules, create and enable pipelines |
| Jobs | Follow jobs and read their logs, start scan, discover, rule and load jobs, cancel, restart or reschedule them, wait for one to finish |
| Variables, dictionary and API keys | Manage project variables, clear the dictionary, manage project API keys |
| Audit | Search the events to see who changed what |
| Maintenance | Run and check database migrations, check the server processes and restart the server (only with the local server) |
The assistant also has an administration guide built in, with step-by-step workflows (for example how to triage a failing job or how to onboard a datasource), so it follows the same procedure every time.
Secrets are never handed to the assistant. Passwords, API keys, OAuth secrets and LDAP bind passwords are replaced by <redacted> in everything it reads.
Two ways to connect
| Remote | Local | |
|---|---|---|
| How it works | The Gigantics server itself answers MCP requests at https://your-instance/mcp | The gig mcp command runs on your computer and talks to Gigantics |
| You need | A Root user to create an MCP API key in Global Settings | The gig program and an email and password |
| The assistant signs in with | The MCP API key | The email and password of a Gigantics user |
| Works when sign-in is LDAP or SAML | Yes | No, it needs a local email and password account |
| Maintenance tools (migrations, server status and restart) | Not available | Available when run on the instance host |
| Best for | Teams, hosted or remote assistants, nothing to install | A single administrator working on the instance machine |
Start with the one that fits you:
Both give the assistant the same tools and follow the same approval rules.
How it keeps you in control
You ask → The assistant calls a tool → Gigantics checks the account's roles → Done
│
└─ destructive tool? → dry run first → a person approves → Done- Permissions: every call is authorized by the Gigantics server using the roles of the account the assistant acts as. A tool description never grants more access.
- Dry run first: a destructive tool (delete a project, remove a user, change authentication, start a load into a real system, and so on) first shows what it would change and does nothing.
- A person approves: you approve in the assistant, or run the same operation yourself. See Approvals and safety.
- Key activity is recorded: creating and deleting MCP API keys appears in the Events, and you can ask the assistant to search them.