Fly.io MCP

Fly.io ships its MCP server inside flyctl, the same command-line tool developers already use to deploy and operate apps on the platform. Starting the server exposes flyctl operations as MCP tools, so an assistant can create and destroy apps, start, stop and scale machines, inspect volumes, read status and logs, set and unset secrets, and check regions and organization details without anyone typing commands into a terminal. It reuses the credentials from an existing flyctl login, so the agent inherits exactly the permissions of the account it runs under. The same binary also acts as a bridge for other MCP servers. It can wrap a local stdio server and put it behind Fly Proxy over HTTP with authentication, or launch a containerized MCP server as a Fly app. That matters when several developers or a hosted agent need to share one server instead of each running a private copy on a laptop. The audience is developers and platform teams already running infrastructure on Fly who want an agent to handle routine operations: spinning up a preview environment, restarting a stuck machine, rotating a secret, or answering questions about what is deployed where. Since destructive commands are reachable through the same interface as read operations, scoping the token the agent authenticates with is the main safety control.