discli

Discord CLI for AI agents and humans

Talk to Discord from your terminal. Send messages, manage channels, run polls, watch live events, and build bots: all with one command-line tool that AI agents can drive too.

Quick look

# Send a message
discli message send "#general" "Hello world"

# React, edit, delete, all from the terminal
discli reaction add "#general" 12345 "👍"
discli message edit "#general" 12345 "Hello, friends"

# Watch events as JSON, pipe into anything
discli --json listen --events messages

# Run a persistent bot driven by an AI agent
discli serve

Every command has a --json flag for scripts. Identifiers accept both names and IDs (#general or 123456789, alice or her snowflake). One token, one CLI, no boilerplate.

What you can do

• Messages: send, edit, delete, search, history, embeds, attachments, replies
• Channels & roles: create, edit, delete, list, get info
• Members: list, info, kick, ban, role assignment
• Threads: create, list, archive, message inside
• DMs: send and read direct messages
• Reactions & polls: manage emoji reactions, run polls with multiple choices
• Live events: discli listen streams every message, reaction, member event as JSONL
• AI agents: discli serve keeps a persistent bot open over stdin/stdout JSONL; drive it from Claude, OpenAI, LangChain, bash
• Interactive components: modals, multi-step workflows with state, persistent dashboards
• Voice (optional extra): join voice channels, transcribe speech, TTS, audio playback, live meeting summaries
• Doctor: discli doctor verifies your setup in one command

Install

pip install discord-cli-agent

Requires Python 3.10+. That's it for text features. No voice deps are pulled in unless you ask for them.

Voice (optional):

pip install 'discord-cli-agent[voice,deepgram]'

You'll also need libopus (apt install libopus0 / brew install opus) and ffmpeg for playback. See docs/guides/voice.mdx for the rest.

Setup

1. Create a bot at Discord Developer Portal.

2. Enable the intents you need (Message Content for reading messages, Members for member lookups, Voice State if you'll touch voice). Pick least-privilege; don't enable everything.

3. Invite the bot to your server with the permissions you actually need.

4. Save your token:

   discli config set token YOUR_BOT_TOKEN
   

5. Verify everything works:

   discli doctor
   

   Doctor reports what's set up and what's missing. Optional features you haven't asked for stay silent.

A bash agent in 8 lines

No Python required, just jq and discli:

discli --json listen --events messages | while read -r event; do
  mentions_bot=$(echo "$event" | jq -r '.mentions_bot')
  if [ "$mentions_bot" = "true" ]; then
    channel=$(echo "$event" | jq -r '.channel_id')
    msg=$(echo "$event" | jq -r '.message_id')
    discli message reply "$channel" "$msg" "Hello! How can I help?"
  fi
done

Pipe discli --json listen into any language. For richer bots, discli serve keeps a single Discord connection open and accepts 54 JSONL actions over stdin. See Serve Mode.

Drive it with Claude

The bundled examples/ai_serve_agent.py is a Claude agent that fully controls Discord: messages, embeds, buttons, selects, modals, channels, roles, polls, threads. @-mention the bot and ask in plain English:

@bot send a blue embed with title "Status" and fields Online=42, Messages=1337
@bot send 3 buttons: Accept (green), Decline (red), Maybe (grey)
@bot create a channel called announcements
@bot create a poll: "Best language?" with Python, Rust, Go

Security & permissions

Four profiles control what an invocation can do:

discli --profile chat message send "#general" "hi"     # ok
discli --profile chat channel delete announcements      # denied

Every destructive action is logged to ~/.discli/audit.log. View with discli audit show. Built-in rate limiter (5 calls per 5s on destructive ops) keeps you from getting banned. Full security model: docs/architecture/security-model.mdx.

Examples

The full reference agents/discord-agent.md drops into any LLM's system prompt: Claude, OpenAI, LangChain, anything that takes a prompt.

Voice (optional)

discli also handles voice channels: joining, speaking via TTS, transcribing speech with streaming STT, and live meeting summaries.

discli voice join "general"
discli voice speak "joining the call now"
discli voice listen        # prints transcripts as people speak

Voice needs the [voice] extra plus a provider ([deepgram] / [elevenlabs] / [openai-voice]). Full walkthrough, including the DAVE encryption fix that makes listening work on modern Discord, lives in docs/guides/voice.mdx.

Configuration

discli config set token YOUR_TOKEN
discli config show

Stored at ~/.discli/config.json. Token resolution: --token flag → DISCORD_BOT_TOKEN env var → config file.

Documentation

Full docs live in docs/:

• Installation · Quickstart · Configuration
• Guides: CLI Usage · Serve Mode · Components & Modals · Voice · Building Agents
• Reference: CLI Commands · Serve Actions · Serve Events · Permission Profiles
• Troubleshooting

Contributing

See CONTRIBUTING.md. Tests run via uv run pytest tests/. Commits follow conventional commits (feat:, fix:, docs:, chore:).

License

MIT. See LICENSE.