Deploy from Cursor

Setup

1. Sign in and create a deploy-scoped token on API Keys (the page generates the block below with your token already in it).
2. Add it to ~/.cursor/mcp.json — or Cursor Settings → MCP → Add server. Windsurf: ~/.codeium/windsurf/mcp_config.json.
3. Reload the editor. The snapdeploy tools appear in the agent's tool list.

{
  "mcpServers": {
    "snapdeploy": {
      "command": "npx",
      "args": ["-y", "@snapdeploy/mcp"],
      "env": { "SNAPDEPLOY_API_KEY": "sd_pat_your_token" }
    }
  }
}

Your repo needs to be on GitHub and connected to SnapDeploy once; the connector returns the connect link if it isn't.

Then, in agent mode

On a failed build the reply includes the error code, a readable message and the last lines of the build log — enough for the agent to fix the code and deploy again. Each attempt counts toward the free deploy limit, and the reply says how many are left.

What the agent actually does

1. Reads the tool list. When Cursor starts, it launches the connector (npx -y @snapdeploy/mcp) as a local process and asks it which tools it offers. The SnapDeploy tools appear in agent mode alongside Cursor's own.
2. Scans before it deploys. A good deploy starts with detect_env_vars: it finds the variables the code reads, flags the required and secret ones and tells the agent where each value lives. The agent asks you for what it cannot know.
3. One call creates, links and builds. deploy creates the container, links the GitHub repository and starts the first build. If linking fails the container is removed again — no half-finished state to clean up later.
4. Builds run on SnapDeploy, not in your editor. AWS CodeBuild with framework detection (Node, Python, Go, Java, PHP, Ruby, static; Dockerfile honoured), then an isolated AWS Fargate task with free SSL. Two to six minutes typically, sixty at most.
5. The agent gets the result back in the chat. The live URL, or the error code, a readable message and the build-log tail — which is exactly what the agent needs to fix the code and try again.

Where the config goes — Cursor, Windsurf, VS Code and friends

The connector needs Node.js 18 or newer on the machine that runs the editor; npx -y fetches and caches the package the first time.

The 15 tools, and the scope each one needs

Scopes nest: read ⊂ deploy ⊂ manage. A token can only reach the rows at or below its scope; the server checks this before any controller runs, and answers 403 INSUFFICIENT_SCOPE with the scope it wanted.

Tokens: scoped, expiring, revocable

The three ways this goes wrong, and what stops each

Troubleshooting

What it costs

The connector itself is free and open. Managed databases and custom domains are available on every tier. India pays in INR (UPI, cards, net banking); everywhere else via PayPal. Full pricing.

Further reading

• SnapDeploy MCP reference — every tool, scopes, scripting the same API with curl, troubleshooting
• Why an agent that can deploy must not be able to delete — the design write-up
• Model Context Protocol — the open standard the connector implements, and its list of client applications
• Cursor documentation and Windsurf documentation — MCP configuration for each editor

Frequently asked questions

Other assistants

Claude Code · Codex CLI and Gemini CLI · Claude Desktop, Cline, Zed and VS Code agent mode use the same JSON block. Building in Lovable? Your export deploys first-try without a connector.