Multi-Agent Task Assistant with A2A & MCP-inspired Architecture

What is Multi-Agent Task Assistant with A2A & MCP-inspired Architecture?

A multi-agent system for article drafting and web research, built with Python, Langchain, Anthropic LLMs, FastAPI, and Docker. It features a custom Agent-to-Agent (A2A) protocol for inter-agent coordination and a Model Context Protocol (MCP)-inspired architecture for interacting with external tool servers.

How to use Multi-Agent Task Assistant with A2A & MCP-inspired Architecture?

Install Docker and Docker Compose, obtain an Anthropic API key, and set up a Google Cloud Project with a storage bucket and service account key. Clone the repository, create a .env file with your credentials, place google_credentials.json in the project root, then run docker-compose up --build. The Streamlit UI is accessible at http://localhost:8501.

Key features of Multi-Agent Task Assistant with A2A & MCP-inspired Architecture

• Article drafting with text generation and cloud storage upload.
• Web research returning summarized findings.
• Supervisor LLM validates, sanitizes, and translates user inputs.
• Custom Agent-to-Agent (A2A) communication protocol over HTTP.
• MCP-inspired tool servers for LLM, cloud storage, and web search.
• Asynchronous task processing with status polling.
• Fully containerized microservice architecture using Docker Compose.

Use cases of Multi-Agent Task Assistant with A2A & MCP-inspired Architecture

• Generate an article draft on a given topic and style, then save it to Google Cloud Storage and retrieve a public URL.
• Perform simulated web research on a topic and receive a list of summarized findings.

FAQ from Multi-Agent Task Assistant with A2A & MCP-inspired Architecture

What LLM models are used and where?

The system uses several Anthropic Claude models: claude-3-7-sonnet-20250219 for the Supervisor LLM (input validation), claude-3-haiku-20240307 for the Article Draft and Researcher agents, claude-3-5-sonnet-20241022 for the Creative LLM MCP Server (text generation), and claude-3-haiku-20240307 for the Web Search MCP Server (simulated search). The Task Manager Agent and Cloud Storage MCP Server do not directly use LLMs.

What are the prerequisites to run the system?

Docker and Docker Compose must be installed. You need an Anthropic API key, a Google Cloud Project with the Cloud Storage API enabled, a GCS bucket, and a service account key file (google_credentials.json) with write permissions to that bucket.

How does the system validate user input?

User inputs are first processed by the Supervisor LLM (Claude model), which validates safety, checks for harmful content, identifies the input language, and translates the topic and style to English before task assignment.

Is the web research real or simulated?

The web research is simulated. The Web Search MCP Server, powered by claude-3-haiku-20240307, generates search results based on a query without connecting to live search engines.

What external services and authentication are required?

Anthropic API (API key via ANTHROPIC_API_KEY environment variable) and Google Cloud Storage (service account key file google_credentials.json, plus GOOGLE_CLOUD_PROJECT and CLOUD_STORAGE_BUCKET_NAME in .env). Internal communication uses HTTP with no additional authentication.