We're excited to announce the official Suga MCP server. Point your agent at https://dashboard.suga.app/api/mcp and it can build and manage your applications without you ever leaving your editor.
Complete platform coverage
Building your project
Create a project and an environment, add containers and volumes, expose a public HTTPS port, set environment variables and secrets, connect a GitHub repository for source builds, add custom domains, set scaling and resource limits, the list goes on. Anything you'd do to shape an environment, you can do from your editor. This makes building out prototypes or connecting existing projects extremely simple.
Monitoring your project
Once your project is running an agent can check deploy status, view application logs, CPU and memory metrics, and pending DNS records for your custom domains. An agent could be set up to detect a failed deployment, read the logs, and open a PR with fixes for you.
Managing your project
Bumping image tags, rotating secrets, adding domains to services. These are all tasks your agent can manage for you after you've gotten up and running.
What about deploys?
We built Suga with the philosophy that truly destructive actions should be difficult to do. That's why on Suga every change is drafted before you apply anything, and when you do there's a complete history of deployments that you can roll back to. Giving an agent the power to apply changes and destroy environments without a human in the loop was directly against this philosophy.
When you're making changes with your agent, it's hard to see what's actually happening when all you get is a tool name and some JSON. That's why we've built the MCP so the agent can work through everything in one pass, then give you a link to the environment, where you can verify pending changes on the canvas and click Apply yourself. You validate the change by looking at it, rather than scrolling back through what your agent did.
This doesn't slow the agent down. It isn't an approval prompt on every change. The agent can make twenty edits without stopping to ask, then you can verify the whole thing once it's done.
What an agent can't reach
Deploys aren't the only destructive action. Deleting a project or an environment tears down running resources, so those sit on the apply side too. There are no delete_project or delete_environment tools, by design.
Secrets are write-only. An agent can set one and rotate one, and it can list which secrets exist on a container, but it can't read a value back out. Connecting an agent is never a way to pull credentials out of your account.
Everything runs as you. The connection uses your existing organization membership and role, so an agent can only reach the orgs you can reach, and can only do what you're already allowed to do.
This replaces the in-app assistant
We've retired the code assistant that lived in the dashboard. It was solving the wrong half of the problem: it knew about Suga but nothing about your codebase, while the agent you already use every day knows your codebase and, until now, nothing about Suga. Meeting you in your editor is the better trade.
Get connected
Claude Code, from your terminal:
claude mcp add --transport http suga https://dashboard.suga.app/api/mcp
Cursor, in ~/.cursor/mcp.json for all projects or .cursor/mcp.json for one:
{
"mcpServers": {
"suga": {
"url": "https://dashboard.suga.app/api/mcp"
}
}
}
Codex, in ~/.codex/config.toml:
[mcp_servers.suga]
url = "https://dashboard.suga.app/api/mcp"
Then codex mcp login suga to authenticate.
Any MCP client that speaks streamable HTTP and OAuth will work. The full tool reference is in the docs. If there's a tool you want that isn't there or an issue you're experiencing, come tell us in Discord.