My Context Progress - A Psychologists Journey to Implementing an MCP
summary
Storyblok builds an MCP server that maps ~140–160 management endpoints into a curated toolset using open API overlays so the source spec stays untouched, and applies a search‑describe‑execute pattern plus response fields to cut context bloat and token use dramatically; governance separates read only, mutating, and destructive actions, CLI remains for deterministic changes, OAuth and pluginified skills are planned, and the implementation is live and working for content analysis, migrations, and content hygiene.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 3:27What is MCP
- 7:22MCP vs CLI and skills
- 9:07Building an MCP server
- 13:42Why not use APIs
- 15:48OpenAPI overlays
- 19:59Progressive discovery
- 21:43Governance and responses
- 24:53Installing skills
- 26:36MCP future and productivity
- 27:41Summary
related talks
23mMCP = Mega Context Problem
37mLetting AI Interface with Your App with MCPs
11mBuilding an MCP Server for a Component
31mMCP support in Envoy
55mCharting the uncharted: Discovering and evolving bounded contexts
21mBuilding Agents (the hard parts!)
01h 01mBounded Contexts: Manage the Understandability of Your Systems
41mExploring MACH Principles