The Real Python Podcast Dangers of Automatically Converting a REST API to MCP
9 snips
Sep 19, 2025 Kyle Stratis, a machine learning consultant and author of 'AI Agents with MCP,' discusses the pitfalls of converting REST APIs to Model Context Protocol (MCP). He emphasizes the importance of avoiding automated tools that compromise an AI agent's context. Steven Gruppetta joins to share insights on Python learning outcomes for beginners. The conversation covers the contrasting interaction models of REST and MCP, the significance of well-defined workflows for agents, and the potential of memory-enabled agents to enhance creativity and developer productivity.
AI Snips
Chapters
Books
Transcript
Episode notes
Actions Not Resources
- MCP centers on actions (RPC/function calls) while REST centers on resources (data access).
- Treating REST endpoints as many MCP tools creates an impedance mismatch and bloated agent context.
Token And Decision Overhead
- Every MCP tool adds tokens and decision-making options to the model's context.
- Excess tools increase token cost and force the model to choose among many functions, harming reliability.
Consolidate Tools Into Single Actions
- Reduce the number of tools you expose to an agent by combining related REST calls into single MCP actions.
- Consolidate server-side logic so agents make fewer round trips and lower cost and error rates.
