The Basics
Core concepts of Ink — the first infrastructure platform built for AI agents. Projects, services, deployments, and how they fit together.
Ink MCP is the first infrastructure platform built for AI agents. Your agent connects via MCP, writes code, and deploys it to production — no DevOps, no configuration files, no CI pipelines. Your agent is the operator.
How it works
- Your AI agent connects to Ink via MCP (Model Context Protocol)
- The agent writes code and pushes it to a git repository
- The agent calls MCP to deploy the code from a GitHub repo or Ink managed git
- Ink auto-detects the language/framework and builds the app
- The app is deployed and gets a live URL at
your-app.ml.ink - The agent monitors metrics, reads logs, and iterates
Every step is an MCP tool call. The agent deploys, observes, and manages — you ship faster.
Key concepts
Workspaces
A workspace is the top-level container. Each workspace has its own billing and usage tracking. Agents can collaborate with each other by sharing a workspace — any agent with access to the workspace can deploy, manage, and observe all projects within it.
Projects
A project is a logical grouping of related services within a workspace. When you sign up, a default project is created. Your agent creates additional projects automatically when deploying with create_service.
Diagram: A workspace contains two projects. Project 1 has Service 1 (React frontend), Service 2 (Node.js API), Service 3 (Worker), and a Resource (SQLite database connected to Service 2). Project 2 has Service 1 (Next.js dashboard), Service 2 (Python API), and a Resource (SQLite database connected to Service 2). Workspaces group projects, projects group services and resources.
Services
A service is a single deployed application. Each service has:
- A git repository (Ink managed or GitHub)
- A build configuration (auto-detected or custom)
- A live URL (default
*.ml.inkor custom domain) - Environment variables for configuration
- Resource allocation (memory and vCPU)
- Metrics (CPU, memory, network) readable by your agent
- Logs (build and runtime) accessible via MCP
Deployments
Every time code is pushed, Ink creates a new deployment. Deployments go through these stages:
- Queued — waiting to build
- Building — auto-detecting framework, installing dependencies, compiling
- Deploying — rolling out to infrastructure
- Active — live and serving traffic
Previous deployments are automatically superseded when a new one goes live. Your agent can check deployment status and read build logs through the get_service MCP tool.
Agent keys
An agent key is an API credential that authenticates your agent with Ink. Keys use the dk_live_ prefix and can be revoked at any time. Each agent should have its own key for security and auditability.
Resources
Resources are managed infrastructure like databases. Your agent provisions them via MCP tools and gets connection credentials back immediately. Currently supported: SQLite via Turso.
Build packs
Ink auto-detects your framework and builds accordingly. Four build strategies are available:
| Build pack | Use case |
|---|---|
| Auto-detect (default) | Automatically detects Node.js, Python, Go, Ruby, Java, Rust, PHP, Elixir, and 25+ other frameworks |
| Dockerfile | Use your own Dockerfile for full control |
| Static | Serve static files directly (HTML, CSS, JS) via nginx |
Default resource allocation
| Setting | Default | Range |
|---|---|---|
| Memory | 256 MB | 256 MB – 4 GB |
| vCPU | 0.5 | 0.5 – 4 |
| Port | 3000 | Any |
| Branch | main | Any |
| Build pack | Auto-detect | railpack, dockerfile, static |