We are the team behind Shuttle (YC S20). We just shipped Neptune, (https://github.com/shuttle-hq/neptune-mcp) an open source MCP server that lets coding agents actually deploy software, not just write code. Behind it is an app deployment platform we built to be friendly for AI agents.

Neptune runs in your IDE. It reads your codebase, analyzes its structure and intent using ASTs and inference, and generates a deterministic infrastructure spec called neptune.json. You can inspect every change, approve it, and apply it. Think: coding agents that can actually ship safely to AWS.

How it works in practice:

1. Take any of your existing apps 2. Install Neptune mcp & login 3. Tell your agent to deploy 4. Watch it do everything else for you and (super important) it keeps you in the loop at every step, asking for your approval and guidance. We’ve spent most of our time implementing guardrails to make this as safe as if you were doing it yourself.

We built this after seeing teams rely on AI to write services, but still hand-craft Terraform and YAML. The AI could write code, but it could not understand the system around it. Neptune tries to close that gap.

Right now we have support for the most basic AWS products (ECS, S3, SecretManager), and we support postgres (thanks to Neon) and we are adding more resources as we speak. For now, your apps are all deployed to our a shared tenancy AWS account, and “bring your own cloud” account is coming soon too.

If you are also excited about a future where coding agents can help you reliably ship stuff into production, try it and share what you think.

If you have tried it on shared tenancy and would like to run it in your own AWS account, just get in touch and we’d be happy to help.