Stripe Projects: Agents Can Now Provision Their Own Stack

Every autonomous agent eventually hits the same wall.

The code isn't the wall. Signing up for services is.

The agent can write features and ship fixes. What it can't do is open a browser, navigate to Clerk's dashboard, generate an API key, and paste it somewhere secure. That step — ten minutes per service, six services per project — always required a human.

Stripe just changed that.

What Stripe Projects is

It's a CLI plugin. Install it, initialize a project, add services with single commands. Supabase, Vercel, Clerk, Neon, Railway, PostHog, PlanetScale, Turso, Chroma, Runloop — all provisionable from the terminal. Credentials go into a vault and sync to your .env automatically.

stripe plugin install projects
stripe projects init my-app
stripe projects add supabase/supabase:free
stripe projects env --pull

No browser tabs, no navigating to API key settings, no pasting secrets into a chat session that lives in plain text on your hard drive.

For a developer working in Cursor or Claude Code, this saves a couple of real hours per project.

But that's not why it matters.

The ceiling that just got removed

Coding agents today have a hard operational limit. They can write code, fix bugs, run deployments. The credential provisioning step — signing up for services, generating API keys, storing them securely — has always been the one thing requiring a human in the loop.

Stripe Projects is explicitly designed to change that. The CLI is deterministic, auditable, scriptable. An agent can run stripe projects add the same way it runs npm install. The credential handoff that always bounced back to a human is now just another command.

That's a capability unlock.

What Stripe is actually building

This isn't Stripe's first move in this direction. They launched the Machine Payments Protocol — an open standard for agents to make payments autonomously. They built Minions, an internal system generating over 1,300 pull requests per week, all agent-produced. Now they're building the credential provisioning layer.

These aren't isolated features — they're the same thesis: agents are going to need to do everything a developer does. For that to work, every step that currently requires a human has to become a command.

Credential generation was one of the last ones standing.

Why this compounds

The AWS IAM analogy is worth sitting with. When AWS launched IAM roles, most developers thought: okay, cleaner permission management. Nice. Then one day they realized every serious cloud architecture depended on it. The infrastructure that seems boring when it ships becomes invisible and essential.

Stripe is doing the same thing with agentic infrastructure — building what agents need to exist, not the agents themselves.

When autonomous systems are everywhere, and they will be, the question won't be whether agents can code. It'll be whether they have the credential layer to operate independently.

Stripe answered that question today.