g4f Discord Bot

A Discord bot powered by gpt4free (g4f) that lets your server chat with AI models — no API keys required.

Features

🤖 /ask — one-shot questions (no history kept)
💬 /chat — conversational mode with per-user message history
🧹 /clear — reset your conversation history
🏷️ /model — show the currently configured model
🔧 /tools — list, enable, and disable MCP tools
🛠️ MCP tool-calling — the AI can autonomously call tools (web search, web scraping, image generation, text-to-audio, and more) via g4f's built-in MCP server. The bot executes the tool, feeds the result back, and loops until the AI has a final answer.
⚡ Streaming responses — edits the message in-place for a live "typing" effect
🔒 Per-user history isolation with configurable length
📡 Live activity feed — an optional channel that mirrors g4f activity in real time: image thumbnails, tool calls, file edits, heavy token usage, server errors, new g4f.dev users, and periodic summaries.

Setup

1. Create a Discord application

Go to the Discord Developer Portal.
Create a new application → Bot tab → Reset Token to get your token.
Enable Message Content Intent under Privileged Gateway Intents.
Invite the bot to your server using the OAuth2 → URL Generator (scopes: bot, applications.commands; permissions: Send Messages, Read Message History).

2. Configure environment

cd projects/discord-bot
cp .env.example .env
# Edit .env and paste your DISCORD_TOKEN

3. Install dependencies

The bot needs discord.py and python-dotenv on top of g4f's requirements:

pip install discord.py python-dotenv

4. Run

python bot.py

You should see:

[INFO] Logged in as YourBot#1234 (id=...)
[INFO] Synced 4 slash commands

Configuration

All settings live in .env:

Variable	Default	Description
`DISCORD_TOKEN`	(required)	Your Discord bot token
`G4F_MODEL`	`gpt-4o-mini`	Model name passed to g4f
`G4F_SYSTEM_PROMPT`	(see .env.example)	System prompt for the assistant
`G4F_MAX_HISTORY`	`12`	Max messages stored per user
`G4F_PROXY`	(none)	Optional proxy for g4f requests
`G4F_ENABLED_TOOLS`	(safe set)	Comma-separated MCP tools to enable at startup
`G4F_MAX_TOOL_LOOPS`	`4`	Max tool-calling rounds before forcing a final answer

MCP tools

The bot integrates g4f's built-in MCP server so the AI can call tools autonomously during a conversation. The flow:

You ask a question (e.g. "What's the latest news on X?").
The model decides to call web_search and returns a tool call.
The bot executes the tool via MCPServer, appends the result to the conversation, and asks the model again.
The loop repeats until the model produces a final answer (or G4F_MAX_TOOL_LOOPS is hit).

Available tools

Tool	Description	Enabled by default?
`web_search`	Search the web via DuckDuckGo	✅
`web_scrape`	Extract text content from a URL	✅
`mark_it_down`	Convert a URL to markdown	✅
`text_to_audio`	Generate an audio URL from text	✅
`image_generation`	Generate an image from a prompt	✅
`python_execute`	Run Python in a sandboxed environment	❌
`apply_patch`	Apply a unified diff patch	❌
`file_read`	Read a file from `~/.g4f/workspace`	❌
`file_read_lines`	Read a line range from a workspace file	❌
`file_search`	Search files in the workspace	❌
`file_write`	Write a file to the workspace	❌
`file_list`	List workspace files	❌
`file_delete`	Delete a workspace file	❌

File/Python/patch tools are disabled by default because they operate on the bot's local filesystem. Enable them only if you trust your Discord users.

Managing tools at runtime

Use the /tools slash command:

/tools                    # list enabled and available tools
/tools action:enable name:python_execute
/tools action:disable name:image_generation

You can also set the startup set via G4F_ENABLED_TOOLS in .env:

G4F_ENABLED_TOOLS=web_search,web_scrape,image_generation

Disabling tools per request

Both /ask and /chat accept an optional tools boolean (defaults to true):

/ask question:"What is 2+2?" tools:False

Changing the provider

bot.py imports OpenaiChat as the default provider. To use a different one, edit the import and the AsyncClient constructor:

from g4f.Provider import Gemini, OpenaiChat, BingCreateImages

client = AsyncClient(provider=Gemini)

See all available providers with:

g4f --help

Live activity feed

The bot can mirror g4f activity into a dedicated Discord channel in real time. Enable it by setting G4F_LIVE_FEED_CHANNEL to a channel ID in your .env.

What gets posted

Event	Trigger	Example
🖼️ Image Generated	Any `/v1/images/generate` or `/v1/media/generate` request	Embed with the thumbnail + full-size link
🔧 Tool Calls	A chat completion whose response includes `tool_calls`	Lists tool names, model, prompt snippet
📝 File Edit	A tool call to `apply_patch`, `file_write`, or `file_delete`	Highlighted separately from other tools
⚡ Heavy Token Usage	A completion using ≥ `G4F_HEAVY_TOKEN_THRESHOLD` tokens	Shows prompt/completion/total token counts
🚨 Server Error	Any request returning a `5xx` status	Path, status, duration
👋 New g4f.dev User	A new user appears in `/members/api/recent-users`	Username, provider, tier, avatar
📊 Activity Summary	Every `G4F_FEED_SUMMARY_INTERVAL` seconds	Rolling counts + top models/providers

How it works

The LiveFeed cog (in live_feed.py) polls two sources on a configurable interval (default 15 s):

{G4F_API_BASE}/api/logs — the g4f API server's request log. The cog remembers the last seen log id and only processes new entries. Image URLs pointing at /media/ or /images/ are rewritten to /thumbnail/ (using G4F_PUBLIC_BASE) so Discord can fetch compact previews.
{G4F_MEMBERS_BASE}/members/api/recent-users — a public endpoint on the g4f.dev members worker that returns the most recently created users. The cog tracks seen provider:username keys and announces new ones.

To keep the channel readable, at most G4F_FEED_MAX_POSTS_PER_CYCLE embeds are posted per poll cycle (additional events are still counted toward the periodic summary).

Setup

Create a dedicated channel in your Discord server (e.g. #g4f-live).
Copy its channel ID (right-click → Copy ID, with Developer Mode enabled).
Add to .env:

G4F_LIVE_FEED_CHANNEL=123456789012345678
G4F_API_BASE=http://localhost:8080          # where the g4f API runs
G4F_PUBLIC_BASE=https://your-public-host     # optional, for Discord-accessible image links
G4F_MEMBERS_BASE=https://g4f.dev            # set empty to disable new-user posts

Restart the bot. You should see Live feed cog loaded → channel ... in the logs.

Configuration reference

Variable	Default	Description
`G4F_LIVE_FEED_CHANNEL`	(unset)	Discord channel ID for the feed. Unset = disabled.
`G4F_API_BASE`	`http://localhost:8080`	g4f API base URL (must expose `/api/logs`).
`G4F_PUBLIC_BASE`	= `G4F_API_BASE`	Public base URL for Discord-accessible image/thumbnail links.
`G4F_MEMBERS_BASE`	`https://g4f.dev`	g4f.dev base URL for new-user posts. Empty = disabled.
`G4F_FEED_POLL_INTERVAL`	`15`	Seconds between polls.
`G4F_HEAVY_TOKEN_THRESHOLD`	`10000`	Token count that flags a completion as "heavy".
`G4F_FEED_SUMMARY_INTERVAL`	`3600`	Seconds between activity summaries.
`G4F_FEED_MAX_POSTS_PER_CYCLE`	`5`	Max embeds per poll cycle (anti-spam).

Project structure

projects/discord-bot/
├── bot.py          # Main bot logic (commands, tool-calling loop)
├── live_feed.py    # Live activity feed cog (image/tool/token/new-user events)
├── mcp_tools.py    # MCP tool manager (definitions, execution, display)
├── .env.example    # Template environment file
└── README.md       # This file

Notes

g4f relies on free third-party providers; availability and quality vary. If a request fails, try a different model or provider.
The bot uses AsyncClient so it stays responsive while streaming.
Discord limits messages to 2000 characters; long replies are truncated.