Fragmented - AI Developer Podcast 302 - MCPs Explained - what they are and when to use them
4 snips
Jan 26, 2026 They unpack what Model Context Protocols are and how they let models plug into external tools and data. They compare MCPs with CLIs and explain server and client roles. They cover real uses like GitHub PR review, Playwright browser control, and instant library docs. They also warn about token bloat, confusion from many tools, and share practical tips for efficient tool use.
AI Snips
Chapters
Transcript
Episode notes
MCP As A Standardized Bridge
- MCP is a standardized protocol that exposes tools via a JSON description so models can access external data and actions.
- It lets models query private, dynamic sources like PRs, Slack, or docs without retraining or leaking data.
CLIs Versus MCPs Tradeoff
- CLIs can often substitute for MCPs since modern models use CLIs well and only consume results back into context.
- MCPs add discoverability and a consistent self-describing interface that CLIs lack, which can justify their use.
Limit MCPs To Avoid Token Bloat
- Avoid adding many MCPs blindly because each tool's definition inflates your model context and burns token budget.
- Keep context tight to reduce hallucinations by limiting tool definitions and choices available to the agent.