GitHub Copilot How to use Fly.io with GitHub Copilot
You want Fly.io inside GitHub Copilot so it can inspect app health and diagnose failed releases, deploy a new image with a safety check, configure a custom domain and verify its certificate, and resize and protect persistent app storage — without you copy-pasting between tabs. Here is the whole setup, once.
Create your Fly.io gateway
The gateway is the thing GitHub Copilot connects to, and it is where you decide which Fly.io tools an agent may use. You create it once and every client reuses it.
-
Create an Arcade account
Sign up and open the dashboard. The free tier is enough to follow this through.
-
Create an MCP Gateway and select your Fly.io tools
In the dashboard, create a gateway and pick the Fly.io tools you want to expose. The gateway is what your client connects to, and it is where tool access is decided.
-
Copy your gateway URL
Every client below needs it.
https://api.arcade.dev/mcp/<YOUR-GATEWAY-SLUG>
Connect the gateway to GitHub Copilot
JetBrains IDEs do not support remote MCP servers with Dynamic Client Registration — configure the gateway as “Arcade Headers” in the dashboard first. Setup also differs on Visual Studio, Eclipse and Xcode; check the docs for yours.
-
In VS Code
Copilot picks up MCP servers configured in VS Code automatically, so follow the VS Code steps and you are done.
-
In JetBrains IDEs
Copilot icon → Open Chat → Agent mode → tools icon → “+ Add More Tools…”, which opens mcp.json.
-
Add the gateway with Arcade Headers
{ "servers": { "mcp-arcade": { "url": "https://api.arcade.dev/mcp/<YOUR-GATEWAY-SLUG>", "requestInit": { "headers": { "Authorization": "Bearer {arcade_api_key}", "Arcade-User-ID": "{arcade_user_id}" } } } } }
Full GitHub Copilot guide in the docs steps verified 2026-08-13
Try these Fly.io prompts
GitHub Copilot can now act on your Fly.io account. Click a prompt to copy it.
Use a credential with only the Fly.io access the agent needs, and require approval before deploying, changing secrets, scaling, or destroying Machines and volumes. Pair it with Contextual Access policies so read-only troubleshooting does not grant permission to make production changes.
