Copilot Discord Bot — GitHub Copilot on the go

Use GitHub Copilot from your phone by chatting in Discord.

You run this bot on a server/PC (home lab, VM, cloud box) that has:

• GitHub Copilot CLI + authentication
• Your repos under a configured root folder (REPOS_ROOT)

Then you use Discord mobile to:

• Create a session (Discord thread)
• Pick a repo with reactions (1..10)
• Approve/deny tool actions with reactions (thumbs up/thumbs down)
• Watch a single "CLI panel" status message update in-place

Demo

Video: media/copilot-discord-bot.mp4

Highlights

• Emoji-first UX: threads + reactions work great on mobile.
• Single status panel: no spam; one message edits in-place.
• Repo sandbox: agent work is scoped under REPOS_ROOT.
• Interactive approvals: prompts show what's being approved.

Security model (read this)

• Set OWNER_DISCORD_USER_ID (recommended). If set, the bot ignores everyone else.
• Keep REPOS_ROOT tight. The bot should only see repos you want it to touch.
• Keep approvals on (cp approve ask) unless you fully trust your tools + repos.
• Optional: AUTO_APPROVE_READ_PERMISSIONS=true reduces prompt spam for read-only actions.

Prereqs

• Node.js 18 or later
  • Verify: node --version should be 18.x or later
• GitHub Copilot CLI installed on the server and authenticated
  • Verify: copilot --version (should not prompt)
  • Authenticate (interactive): run copilot once and complete sign-in
  • Authenticate (service-friendly): set GH_TOKEN / COPILOT_GITHUB_TOKEN
• A Discord application + bot token

Create a Discord bot + invite it

In the Discord Developer Portal:

1. Create an Application
2. Create a Bot and copy the token
3. Enable Privileged Gateway Intents:
   • Message Content Intent (required)
4. Invite the bot to your server (OAuth2 -> URL Generator):
   • Scopes: bot
   • Permissions (minimum recommended):
     • View Channels
     • Send Messages
     • Read Message History
     • Create Public Threads
     • Send Messages in Threads
     • Add Reactions
     • Use External Emojis (optional)

Configure

Copy .env.example to .env and fill in the values:

cp .env.example .env

Required variables:

• DISCORD_BOT_TOKEN — your bot token
• OWNER_DISCORD_USER_ID — your Discord user ID (recommended)
• REPOS_ROOT — root folder containing repos (example: /home/user/git)

All configuration variables (with defaults):

Variable	Default	Description
DISCORD_BOT_TOKEN	(required)	Discord bot token
OWNER_DISCORD_USER_ID	(optional)	Only this user can control the bot
REPOS_ROOT	(required)	Root folder of all repos
DATA_DIR	data	SQLite + Copilot session state directory
DEFAULT_MODEL	gpt-5	Default Copilot model
COPILOT_CLI_PATH	(uses PATH)	Explicit path to Copilot CLI
DEFAULT_AUTO_APPROVE_PERMISSIONS	false	Auto-approve all permission requests
AUTO_APPROVE_READ_PERMISSIONS	true	Auto-approve read-only requests
TURN_TIMEOUT_SECONDS	900	Seconds before a turn times out
DEBUG_PERMISSION_PAYLOAD	false	Show debug info in permission prompts
PORT	5000	HTTP server port
HOST	127.0.0.1	HTTP server host

Run (dev)

From repo root:

npm install
npm run dev

Health endpoint:

• http://localhost:5000/health

Build

npm run build
npm start

Install as a background service (recommended)

These scripts build the app, set environment variables, and register an always-on service.

Windows (Windows Service)

Run from an elevated PowerShell:

powershell -ExecutionPolicy Bypass -File .\scripts\install-service-windows.ps1

Common options:

• -ReposRoot "E:\Git"
• -OwnerDiscordUserId 123456789012345678
• -GhToken "..." (recommended for services)
• -CopilotCliPath "C:\Path\to\copilot.exe"

Uninstall:

powershell -ExecutionPolicy Bypass -File .\scripts\uninstall-service-windows.ps1

Linux (systemd)

Make scripts executable and run with sudo:

chmod +x scripts/*.sh
sudo ./scripts/install-service-linux.sh --user <username> --repos-root /path/to/repos

The installer builds the project, copies output to the install directory, writes an env file at /etc/copilot-discord-bot.env (mode 600) and creates a systemd unit.

Uninstall:

sudo ./scripts/uninstall-service-linux.sh --remove-env

Use it (emoji-first)

Create a session

In any normal text channel:

• cp new my-session

The bot creates a thread (that thread is the session) and posts a repo picker.

Pick a repo

In the session thread:

• React with 1..10 on the repo picker message.

Fallback: cp repo <path>

Chat

Just send normal messages in the thread.

You will see a single status panel updated in-place (state/repo/model/doing).

Change model

In the session thread:

• cp model (shows current)
• cp model gpt-5.2 (sets for this session)

Approve/Deny tool actions

When Copilot requests permission, the bot posts a prompt message.

• React thumbs up to approve
• React thumbs down to deny

Troubleshooting

Copilot runtime failed to start

This usually means the copilot process exited immediately (often because PATH resolved to a VS Code shim that prints an interactive install prompt).

• Run which copilot (Linux/Mac) or where copilot (Windows) and verify it points to the installed copilot binary
• Run copilot --version in the same shell you launch the bot from
• If it prompts to install/reinstall, restart your terminal (PATH update) and try again
• If it still points to a shim, set COPILOT_CLI_PATH to the full binary path

Permission prompt too vague

The Copilot SDK is in preview and permission payload fields can vary. Enable:

• DEBUG_PERMISSION_PAYLOAD=true

This adds a debug panel to permission prompts.

Auto-approve policy (per session)

• cp approve ask (recommended)
• cp approve always

MCP server example

Enable the GitHub MCP server for a session:

cp config set {"model":"gpt-5.2","mcpServers":{"github":{"type":"http","url":"https://api.githubcopilot.com/mcp/","tools":["*"]}}}