Members-Only
Recent Talks & Demos are for members only
You must be an AI Tinkerers active member to view these talks and demos.
AI Workers That Create and Operate Their Own MCP Apps
AI workers autonomously build, deploy, and operate MCP apps, showcasing a config-driven system for dynamic software creation and improvement at scale.
A platform where AI workers (tedis) autonomously build, deploy, and operate MCP apps, both internal tools and customer-facing products as ChatGPT apps. Not “AI-assisted development.” The Tedi IS the developer, the ops engineer, and the product manager.
I’ll walk through the technical build:
-
The MCP app lifecycle: A tedi takes an API endpoint, maps it to MCP tools (config-driven D1 rows, zero handler files), generates widget layouts via genUI, validates output with screenshot-based QA (Cloudflare Browser Rendering → R2), and deploys — all autonomously.
-
131 tools, one universal handler: Every MCP tool is a database row. Adding a tool = inserting a row. No code, no deploy.
-
Persistent memory for improvement: Tedis carry a semantic knowledge graph (Brain MCP: Upstash Vector + D1) so each app they build is better than the last. Patterns, gotchas, and architecture decisions compound across sessions.