🌱 Sprout an Agent with Haystack & MCP

Spring has sprung — flowers are blooming, birds are chirping, and it’s the perfect time for some fresh ideas to take root 🐥

Spring cleaning can wait… you’re here to level up your AI dev skills with Haystack and take part in a blossoming new movement: Anthropic’s Model Context Protocol (MCP) 🌐

This challenge invites you to sprout a Haystack Agent that speaks MCP and connects to external systems — starting with GitHub.

🧠 What’s MCP?

The Model Context Protocol is an emerging open standard that unlocks two-way, secure, and structured communication between your AI applications and the tools they use.

In this model:

MCP Servers act like translators between your app and real-world tools — whether it’s API like GitHub, a database, or a local file system. Think of them as smart wrappers that expose APIs and systems to AI in a consistent, reliable way.
MCP Clients (like your Haystack Agent) handle everything from connection management to calling specific tools, to parsing and acting on responses.

🪻 Your Spring Challenge

You’re going to create a tool-calling Haystack Agent that acts as an MCP Client — capable of connecting to the GitHub MCP Server and performing actions like:

Creating forks
Opening PRs
Reading file content
Updating issues
… or whatever GitHub action you choose to empower it with ✨

🌷 How to Participate

Getting started is easy:

Fork the deepset-ai/spring-into-haystack repo on GitHub
Fill in the missing pieces to build your MCP-connected Agent
Push your code to your repo
Submit your solution so we can see your creation in full bloom!

🔧 You’ll Use:

Agent – component for the smart decision-maker
MCPTool – lets your agent talk to the MCP Server

🎯 Requirements

A GitHub Token with relevant permission