getting an API
roll your own
Wrap the agent in FastAPI or Express, write route handlers, define schemas, and containerize it.
a2a cloud
`a2a deploy` generates a REST/OpenAPI surface from your tools — no hand-written routes.
one command, a production API
Deploy your AI agent as an API, batteries included.
Getting an agent to run is not the hard part. Making it a production service — an API with schemas, a database, authentication, and an audit trail — is. a2a cloud collapses that into one command, with managed Postgres, auth, TLS, and signed receipts for authenticated Agent API and public /invoke executions.
REST · MCP · A2A · frontend — one deploy
0
commands to production
0
surfaces per agent
signed entry points
the problem
Your agent works. Turning it into a service is the other 80%.
A working agent in a notebook is a long way from a service other systems can depend on. To ship it as an API you write route handlers and schemas, provision and migrate a database, integrate authentication, add logging you can't really trust, containerize it, and put TLS in front. None of that is your agent's logic — it's the undifferentiated infrastructure every team rebuilds, badly, before their agent does anything for a caller.
Wrapping the agent in FastAPI or Express and hand-writing routes and schemas is pure boilerplate.
Provisioning Postgres, wiring the connection, and managing migrations is a project of its own.
Building auth — token issuance, verification, rotation — delays v1 by weeks.
Bolt-on logging is mutable and can't prove what the agent did on a given request.
the a2a way
One deploy. The API, the database, the auth, the proof.
a2a cloud makes the whole production surface a single output. You ship the agent; the platform generates the API, provisions the database, authenticates callers, and signs defined execution calls through the trusted control-plane path.
One command, a real APIManaged Postgres, wired inAuth without building authSigned execution callsMCP, A2A, and a frontend tooHosted service + code sandbox
One command, a real API
`a2a deploy` turns your agent into a REST service with a generated OpenAPI spec. Every tool becomes an endpoint with typed request and response schemas — callable from curl, an SDK, or another service.
Managed Postgres, wired in
A managed Postgres database is provisioned and connected on the first deploy. Your agent has durable state from the start, not after you go set up a database and paste a connection string.
Auth without building auth
The API is authenticated by default. Callers present short-lived signed tokens and scoped grants — you don't stand up OAuth, an identity provider, or a key-management scheme to ship v1.
Signed execution calls
Agent API and public /invoke executions receive Ed25519-signed, persisted receipts for observed inputs and terminal results. Only authenticated Agent API paid calls have separate price, fee, and payout rows; public gateway receipts record zero platform economics.
MCP, A2A, and a frontend too
The same tools are published as MCP tools and A2A operations, and you can host a frontend on the same deploy. One agent, reachable as an API, an MCP server, an A2A peer, and a web app.
Hosted service + code sandbox
The agent API runs as a Knative or Kubernetes container service with TLS on a stable hostname. Explicit ctx.sandbox code-execution commands use a separate microsandbox microVM.
side-by-side
Building it yourself vs. `a2a deploy`.
database
roll your own
Provision Postgres, manage migrations, wire the connection string, handle secrets.
a2a cloud
A managed Postgres database is provisioned and connected on the first deploy.
auth
roll your own
Integrate an identity provider or roll token issuance, verification, and rotation yourself.
a2a cloud
Authenticated by default with short-lived signed tokens and scoped grants.
audit
roll your own
Add logging, hope it captures enough, and accept it's mutable and unprovable.
a2a cloud
Agent API and public /invoke executions produce individually signed receipts.
surfaces
roll your own
Rebuild the same logic to also speak MCP, A2A, or serve a UI.
a2a cloud
REST, MCP, A2A, and frontend hosting from one deploy of one agent.
dimension
roll your own
a2a cloud
getting an API
Wrap the agent in FastAPI or Express, write route handlers, define schemas, and containerize it.
`a2a deploy` generates a REST/OpenAPI surface from your tools — no hand-written routes.
database
Provision Postgres, manage migrations, wire the connection string, handle secrets.
A managed Postgres database is provisioned and connected on the first deploy.
auth
Integrate an identity provider or roll token issuance, verification, and rotation yourself.
Authenticated by default with short-lived signed tokens and scoped grants.
audit
Add logging, hope it captures enough, and accept it's mutable and unprovable.
Agent API and public /invoke executions produce individually signed receipts.
surfaces
Rebuild the same logic to also speak MCP, A2A, or serve a UI.
REST, MCP, A2A, and frontend hosting from one deploy of one agent.
questions
Frequently asked.
How do I deploy an AI agent as an API?
Run `a2a deploy` against your agent. a2a cloud generates a REST service with an OpenAPI spec — each tool becomes a typed endpoint — and provisions a managed Postgres database, authentication, TLS, and a stable hostname. Your agent is a callable production API without any of the plumbing written by hand.
What do I get beyond the REST endpoints?
The same skills are exposed through MCP and public /invoke, and you can host a frontend. Agent API, public /invoke, and standard MCP tools/call executions receive receipts; only Agent API paid calls have linked economics. Hosted skills run in containers, while explicit ctx.sandbox commands use a separate microsandbox microVM.
Do I need to write route handlers or an OpenAPI spec?
No. The REST surface and its OpenAPI spec are generated from the tools you already expose on your agent, with typed request and response schemas. You don't hand-write route handlers, and the spec stays in step with your code across deploys.
Is the agent API production-ready or just a demo?
It is a production container service with TLS on a stable hostname, managed Postgres, and authentication on the front door. Defined execution entry points produce signed receipts, while explicit code-execution commands can use the separate microsandbox boundary.
Which agent frameworks can I deploy as an API?
LangGraph, CrewAI, the OpenAI Agents SDK, and custom Python or TypeScript agents all deploy as-is. a2a cloud wraps whatever you bring in the REST, MCP, and A2A surfaces, so you don't rewrite the agent to expose it as an API.
keep reading
Related guides.
All guides live in the guides index.
ship the agent, skip the plumbing
Deploy your agent as an API today.
a2a cloud turns any agent into a production REST/OpenAPI service with managed Postgres, MCP, frontend hosting, auth, and TLS. Trusted control-plane paths sign Agent API calls, public /invoke executions, and standard MCP tools/call executions.