Discord-AIBot

Discord bot for supporting AI/LLM chat applications powered by the Model Context Protocol (MCP), allowing for numerous integrations

For general MCP resources, see Arkestra:cookbook/mcp/README.md

Configuration

cp config/example.main.toml config/main.toml

Then edit main.toml as needed. It specifies your LLM endpoint and model context resources such as MCP servers.

Running

If you included toys.b4a.toml in your main.toml you'll need to have that MCP server running. In a separate terminal cd demo_server then run

uv pip install -Ur requirements.txt
uvicorn toy_mcp_server:create_app --factory --host 127.0.0.1 --port 8902

Make sure you set up any other MCP or other resources you've specified in your B4A. Now you can run the bot.

# Assumes you've exported DISCORD_TOKEN="YOUR_TOKEN"
python mcp_discord_bot.py --discord-token $DISCORD_TOKEN --config-path config

Structlog/rich tracebacks can be elaborate, so there is a --classic-tracebacks option to tame them

For very copious logging you can add --loglevel DEBUG

Note: you can use the environment rather than --discord-token & --config-path

export AIBOT_DISCORD_TOKEN="YOUR_TOKEN"
export AIBOT_DISCORD_CONFIG_PATH="./config"
python mcp_discord_bot.py # Reads from env vars

Implementation notes

• discord.py
• mcp-sse-client MCP client library

Using mlx_lm.server

uv pip install mlx-omni-server
mlx-omni-server --port 1234
# uv pip install mlx mlx_lm
# mlx_lm.server --model mlx-community/Llama-3.2-3B-Instruct-4bit --port 1234

Note: with mlx-omni-server ran into RuntimeError: Failed to generate completion: generate_step() got an unexpected keyword argument 'user'

Fixed with this patch:

diff --git a/chat/mlx/mlx_model.py b/chat/mlx/mlx_model.py
index da7aef5..094ae9c 100644
--- a/chat/mlx/mlx_model.py
+++ b/chat/mlx/mlx_model.py
@@ -45,6 +45,9 @@ class MLXModel(BaseTextModel):
 
     def _get_generation_params(self, request: ChatCompletionRequest) -> Dict[str, Any]:
         params = request.get_extra_params()
+        # Exclude user. See #37
+        if "user" in params:
+            del params["user"]
         known_params = {
             "top_k",
             "min_tokens_to_keep",

There are many local MLX models from which you can pick

Checking MCP servers

For SSE servers, you can check with curl, e.g.

curl -N http://localhost:8901/sse