Are MCPs the New APIs?

Listen to full episode:

Summary: In the first ever episode of Context Window, Anant and Ed explore one of the biggest shifts happening in enterprise AI: how models, agents, and tools are changing the way software connects to data and capabilities. They compare MCP to earlier API-era patterns and dig into what still matters most: good design, clear descriptions, reliability, and trust.

Chapters:

04:54 — Intro: Context Window livestream begins

06:19 — Why MCP and tool use matter for enterprise AI

10:33 — Tools as the new interaction layer: RAG vs. tools vs. richer capabilities

15:53 — What MCP standardizes: API design, tool descriptions, and callable actions

22:43 — Good vs. bad backend design for model use

24:29 — Why context and use-case descriptions matter

28:48 — Atomic capabilities vs. larger agentic tasks

30:12 — Multi-agent / MCP composition challenges

41:00 — Testing MCP servers, agent chains, and the "freeze-dry" pattern

50:46 — Advice for API teams: getting started with MCP, docs, and context

57:50 — Trust, reliability, and what's next

Sound Bites:

“Good API design makes for great tools.”

“Good system design doesn’t go away just because you can do things in a more flexible way.”