AI Agents Provision Their Own Infrastructure: Zero-Human MCP Onboarding
Most "AI-native" infrastructure still has a human in the loop somewhere — someone has to click "create project" and copy an API key before the agent can do anything. ZeroDB's /api/v1/public/instant-db endpoint removes that step entirely: an agent can provision its own database over MCP with zero human involvement.
The decision tree
When an agent calls the public instant-db endpoint, provision.js walks a decision tree: is this a brand-new agent (no existing project), a returning agent (valid claim token), or a suspicious pattern (rate-limit or abuse signal)? Each branch provisions, reuses, or rejects accordingly — without a human reviewing the request.
Atomic credential issuance
A project and its first API key are created in a single atomic write. There's no window where a project exists without credentials, and no window where credentials are issued without a backing project — both failure modes that would otherwise require manual cleanup.
The claim token security model
Because there's no human to authenticate the first request, ZeroDB issues a short-lived claim token scoped to exactly one project. The token proves "I am the agent that just provisioned this," without requiring a password, OAuth flow, or email verification up front. A human can later claim the project permanently through the dashboard, but the agent isn't blocked on that happening.
Why this matters
Agents that can provision their own storage, memory, and vector search without pausing for a human to set up an account can actually complete end-to-end tasks autonomously — spin up, use, and (if needed) tear down infrastructure as part of their own workflow instead of a setup wizard a person has to drive.
Originally published on the AINative Studio blog.