AI Automation

GitHub Package Discovery Agent

Task decomposition

User describes a package need in chat (e.g. “I need a Python sentiment analysis package”).
A config step tags the request with a language hint.
The agent searches GitHub through a connected MCP tool and returns exactly 3 numbered candidates, each with name, description, language, popularity, and a reason it fits.
The agent asks the user to reply with 1, 2, or 3.
On a numeric reply, the agent recalls the earlier candidate list from conversation memory and pulls a deeper summary of the selected repo.
The deeper summary is returned to the user.

Objective of the workflow

This workflow demonstrates an agent connected to an external tool through MCP (Model Context Protocol) rather than a direct API call, letting it search a live external source instead of relying on its own training data.
It also shows a two-step conversational pattern: a broad first pass that narrows options to three, followed by a targeted follow-up that uses memory to resolve a short numeric reply back to a specific candidate. Built for CIS 515 (AI and Data Analytics Strategy).

Architecture