Autonomous MCP Agent

An agent that finds open-source MCP servers on GitHub at run time, runs them in throwaway uvx environments, and replans when a tool fails. By Partha Sarkar Project page: https://towardsdatascience.com/shipai/projects/auto-mcp ShipAI is a Towards Data Science project. You can learn more about it here: https://towardsdatascience.com/shipai/about Key takeaways. 1. A planner splits the task, a discovery agent finds MCP servers on GitHub, and uvx runs each in isolation. 2. When the SQLite insert failed on a missing table, the LangGraph supervisor replanned: create the table first, then retry. 3. Three finance servers failed the stock briefing; a whitelisted yfinance-mcp fixed it, and working tools get reused for later steps. Chapters: 0:00 Why fixed tool sets 0:51 Planning a trip across India 2:58 Stock briefing: three servers fail 4:35 Whitelisting yfinance-mcp 5:15 Self-healing a SQLite insert 8:07 The agent pipeline