staticbot-mcp

MCP server for Staticbot — lets AI agents (Claude Code, Cursor, and any MCP-compatible runtime) orchestrate deployment workflows via the Staticbot API.

What Staticbot does

Staticbot is a managed operations platform for complex, long-running deployment workflows. It hosts static websites and web apps inside your own AWS account using S3 and CloudFront — no vendor lock-in, no PaaS markup, full ownership of the infrastructure.

Core workflows:

• Deploy to AWS — Static websites and web apps to S3 + CloudFront CDN, with custom domains and automatic SSL via ACM
• Supabase migration — Orchestrates moving a full Supabase project (database, edge functions, storage, auth) to your own AWS infrastructure
• Multi-stage deployments — Dev, preview, and production stages per website with independent lifecycle management

Staticbot owns the things agents handle poorly: state that persists across sessions, credentials that should never appear in a context window, approval gates before destructive operations, and accumulated operational knowledge from real-world failure modes.

Prerequisites

• A Staticbot account
• An API key — generate one at Settings → API Keys in the Staticbot UI
• Node.js 18+

Setup

Claude Code

Add to your project's .mcp.json or global MCP config:

{
  "mcpServers": {
    "staticbot": {
      "command": "npx",
      "args": ["-y", "staticbot-mcp"],
      "env": {
        "STATICBOT_API_KEY": "sk-your-api-key-here"
      }
    }
  }
}

Cursor / other MCP clients

Point your MCP client at:

command: npx -y staticbot-mcp
env: STATICBOT_API_KEY=sk-your-api-key-here

Environment variables

Available tools

Migrations

Migration statuses: PENDING → IN_PROGRESS → PAUSED_FOR_APPROVAL → IN_PROGRESS → COMPLETED (or COMPLETED_WITH_ERRORS / FAILED). Migrations may also pause at PAUSED_FOR_USER_ACTION when a human step is required (e.g. DNS record setup).

Deployments

Stacks

How agent-assisted deployment works

Tools return immediately with an ID and a statusUrl — a deep link into the Staticbot UI. The agent gives this to the human, then polls or waits.

Human: "Deploy my site to AWS on example.com"

Agent:
  1. Calls create_stack(...) → gets stack_id + status_url
  2. Tells human: "Deployment started. Track progress here: [status_url]"
  3. Polls get_deployment() until status changes
  4. When status = WAITING:
       Tells human: "DNS setup required. Add these records to your domain
       registrar, then come back and confirm. Details: [status_url]"
  5. Human configures DNS, confirms in Staticbot UI
  6. Agent resumes polling
  7. When status = COMPLETED:
       Tells human the live URL

For migrations:

Human: "Migrate my Lovable project to my own AWS"

Agent:
  1. Calls list_migrations() to check for existing work
  2. Starts migration → gets migration_id + status_url
  3. Shares status_url with human immediately
  4. When status = PAUSED_FOR_APPROVAL:
       "Discovery complete — Staticbot found 3 tables, 2 edge functions,
        1 storage bucket. Review the plan and approve at [status_url]"
  5. Human reviews and approves in the Staticbot UI
  6. Agent calls confirm_migration() or human approves in UI
  7. Polls to completion, reports live URL

The agent handles orchestration. Staticbot handles execution, state, credentials, and oversight. The human has a live view throughout.

What Staticbot is not

• Not a CI/CD platform — it doesn't replace GitHub Actions or deployment pipelines
• Not a no-code builder — it deploys applications, it doesn't create them
• Not a generic infrastructure tool — use Terraform or Pulumi for arbitrary infra; Staticbot is for proven, opinionated playbooks

License

MIT