Just open-sourced the @ice_panel MCP Server! π§π₯
AI Agents can now trace & build your C4 diagrams.
build with mcp, and use browser subagents to capture screenshots & flows video recordings!
Built 100% with @antigravity π https://t.co/RIUOfEjLcU
#MCP#AI#ICEPANEL#C4
@aginaut@ice_panel@antigravity 100%. π― That's the exact workflow this unlocks! You point your agent at your local source code, and it uses the MCP to generate the IcePanel C4 model directly from your files. No manual hand-waving, so the diagram never drifts from the actual codebase.
Just open-sourced the @ice_panel MCP Server! π§π₯
AI Agents can now trace & build your C4 diagrams.
build with mcp, and use browser subagents to capture screenshots & flows video recordings!
Built 100% with @antigravity π https://t.co/RIUOfEjLcU
#MCP#AI#ICEPANEL#C4
Spot on! It's actually highly customizable.
Since the MCP server builds its endpoints via pure AST code-reflection directly from the official IcePanel SDK, it natively covers everything IcePanel offer: Diagrams, Models, Domains, Drafts, Versions, Tags, etc.
You're never locked into a rigid automated layout! You can simply instruct the agents in chat to map the C4 diagram exactly how you visualize it and create animated flows as you see it fit ad hoc, or you could design a custom Agent 'Skill' so the AI permanently adheres to your specific layout preferences automatically! π§π₯
@ice_panel@antigravity Fun fact about this video: I didn't record it! π π
This is a complete screen recording generated natively by Google Antigravity. It used its browser subagent to autonomously navigate the and visually trace the C4 flows that it had just finished building via the MCP server!
@ice_panel@antigravity All of the C4 architecture diagrams attached in the original tweet above were mapped out and generated completely autonomously by an AI agent using this exact MCP server! π₯π€π
@ice_panel@antigravity Finally, you shouldn't be locked into a single proxy environment. π’
Because it's built on Web Standards, you can run it exactly how your workflow demands:
π³ Docker, over standard STDIN
π» Localhost: Fast native execution
βοΈ Cloudflare: Deploy Serverless to the edge globally
@ice_panel@antigravity Because AI workflows require absolute reliability, this MCP server ships with 100% test coverage. π‘οΈ
It uses a E2E validation suite running live against the endpoints to catch network regressions instantly. Before your agent writes code, need to be sure the tool actually works.
@ice_panel@antigravity Let me tell you about how it works! π§¬
Instead of manually maintaining fragile APIs, the server uses TypeScript AST parsing to reflect off the official SDK. When the API updates, the scraper auto generates exact schemas and endpoints. 100% structural parity with zero error.
π Skia ripple shader β three flavors! Swipe to pick your favorite π
1οΈβ£ Clean wave (insp. @JorgeCastilloPr )
2οΈβ£ Subtle prism (personal fave)
3οΈβ£ Full Big rainbow + animated text (IOS effect insp. @AdemarR92 )
All adjustable via Reanimated uniforms in Expo.
Which one are you feeling? Vote!
Tomorrow: ripple breakdown π
#ReactNativeSkia #Skia #Reanimated #ReactNative #Expo #BuildInPublic
@Shopify@swmansion@expo
So I gave in to vibe coding. π
Asked him to make 3d fps using Three.js.
Had around 1h long session with initial plan for made in grok3 and then used cursor and Claude3.7
https://t.co/OAqv4IEZjU
#cursorai#claude#grok3#ai#aigamedev#vibecoding#gamedev#threejs
it's 2025. blows my mind there are devs out there rawdogging code manually without *any* llm support. no cursor, no composer, no chat, no copilot, not even chatgpt