Happy slice of mcp pizzaMCP.pizza

mcp-assistant

MCP.Pizza Chef: AIAtrium

mcp-assistant is an MCP client designed as an AI assistant to help busy users manage context switching efficiently. It leverages the MCP protocol and integrates with Arcade for tool usage without requiring local or remote MCP servers. It supports OpenAI and Anthropic LLMs and offers observability through Langfuse for tracing LLM calls, making it ideal for multitasking workflows.

Unmaintained · No commits in 15 months.

Use This MCP client To

Manage multiple tasks with seamless context switching Integrate LLM-powered tools without local MCP servers Use a single LLM key for OpenAI or Anthropic models Trace and observe LLM calls with Langfuse integration Run an AI assistant in Python virtual environments Simplify multitasking for busy professionals

README

MCP Agent

This is an AI assistant that runs using MCP.

How to install before running

Basic Setup

  • uv:
  • legacy
    • create a python venv and pip install -r requirements.txt.

This application uses Arcade to facilitate tool use so you don't need to run any local MCP servers or connect to any remotely in the cloud. Please sign up for Arcade and get a free api key here once registered to use this repo.

Set the following enviroment variables in an .env file in the project root:

ANTHROPIC_API_KEY=
ARCADE_API_KEY=
OPENAI_API_KEY=

You only need ONE LLM key. This repo only supports OpenAI and Anthropic models currently.

By default the agent will kick off an OAuth flow with Arcade for any tools that its not authorized to use. This will cause the exectuion to pause until that flow completes. If you want to skip any unauthorized tools without kicking off OAuth so that it doesn't block, set SKIP_CLI_AUTH=true.

Observability and tracing

This project uses Langfuse for observability and tracing. Its not mandatory but its helpful to see the dozens of LLM calls that occur nicely formatted in a UI. To utilize it, set these environment variables in the .env:

LANGFUSE_SECRET_KEY=
LANGFUSE_PUBLIC_KEY=
LANGFUSE_HOST="https://cloud.langfuse.com"

How to Run

The MCP Assistant application is heavily customizable. You can do anything email triaging to lead qualification.

The default workflow generates a daily briefing over various tools like email, calendar, whatsapp and Notion that tells you your schedule for the day, gives you background research from the internet about who you're meeting with and tells you what is urgent.

mcp-assistant FAQ

How do I install mcp-assistant?
Create a Python virtual environment, then run `pip install -r requirements.txt` to install dependencies.
Do I need to run local MCP servers to use mcp-assistant?
No, mcp-assistant uses Arcade to facilitate tool use without local or remote MCP servers.
Which LLM providers does mcp-assistant support?
It supports OpenAI and Anthropic models, with plans to support more providers like Claude and Gemini.
How do I configure API keys for mcp-assistant?
Set your API keys for Anthropic, Arcade, and OpenAI in an `.env` file in the project root.
What is the role of Langfuse in mcp-assistant?
Langfuse provides observability and tracing for LLM calls, helping you monitor interactions in a UI.
Can I use mcp-assistant with only one LLM API key?
Yes, you only need one LLM API key from either OpenAI or Anthropic to run the assistant.
Is mcp-assistant suitable for users who frequently switch contexts?
Yes, it is specifically designed to help busy people manage context switching efficiently.