MCP is the New REST
summary
Typeform describes why REST proves awkward for LLM-driven UIs and how Model Context Protocol (MCP) fixes it by exposing semantically typed tools instead of generic JSON patches. Typeform builds an internal Go library to host MCP servers inside each service, runs an MCP gateway that routes by tool name, publishes 60 tools (manifest metadata with descriptions), lazy-loads schemas to save tokens, and replatforms a previous REST-based agent (which served millions of requests) to MCP at the start of the year with automated evals and a feedback form.
AI-written summary · about
chapters (17)
- 0:00Introduction to MCP
- 1:52What is MCP
- 4:20Shift in UX with AI
- 8:30REST problems for AI UX
- 14:51MCP better for LLMs
- 18:48MCP as new API paradigm
- 21:57Scaling MCP in teams
- 28:36Summary
- 31:03Q&A
- 31:25MCP vs REST architecture
- 34:15Token costs with MCP
- 37:30Streaming responses in MCP
- 39:35AI-native text box UX
- 41:23LLM tool behavior consistency
- 44:10Tracking user experience
- 46:00MCP vs REST limits
- 49:55Q&A session concludes
related talks
46mMCP for LLM Agents, Infrastructure APIs & Data Services
23mThe Design of MCP Servers
32mMCP for dummies: Feed more context to LLM
33mThe Model Context Protocol: Connecting AI to Everything
50mBeyond local tools: Dive into the Model Context Protocol (MCP)
35mMCP in Action: Real-World Use Cases & Design Patterns
25mFrom Idea to Integration: An Intro to the Model Context Protocol (MCP)
02h 49mMCP Servers: Best Practices, Design Choices and Their Consequences