These are the questions builders asked us before the Build for SPECS session, with answers from the Snap team. We'll keep this page updated as much as we can. If you have more questions, ask them in the discussion forum.

Set up and build

What's the stack for building with SPECS and AI coding agents?
Lens Studio (TypeScript), the Lens Studio MCP server, and the CLAD agent plugin, which includes skills and agents for Claude Code, Codex, and Cursor.

CLAD won't let me sign in to the developer platform. What do I do?
Use Lens Studio 5.23.2 instead. You don't need a SPECS developer account for CLAD.

How do I connect my coding agent to Lens Studio?
In Lens Studio, go to AI Assistant → AI Model Context Protocol (MCP) → Configure Server → Copy MCP Config, then add it to your agent. You can also start the agent from inside the project folder. Docs: Lens Studio AI → Developer Mode.

What does the workflow look like?
Plan, build the scene and scripts through MCP, check in Preview (logs, screenshots), test on device, then publish.

Any tips for beginners?

  • Start from a SPECS sample.

  • Keep agent conversations short, one feature per conversation.

  • Ask the agent to check its work in Preview.

  • Use the Interaction Kit and UI Kit instead of writing input from scratch.

Where can I see examples?
CLAD docs, video 1, video 2, video 3

Test before you submit

Is there a simulation I can try?
Yes. Lens Studio Preview includes SPECS simulation: a simulated hand, pinch, world mesh, and camera. Interactive Preview lets you move around and interact with a mouse and keyboard.

How do I make sure an experience I build without glasses works well for people wearing them?
Build for the SPECS field of view and keep content in a comfortable range (about 0.5 to 3 m), never head-locked to the face. Use the Interaction Kit for hand and pinch input. Keep scenes light, since this is mobile hardware. Use Preview's SPECS simulation, then test on a real device before you submit.

How do I test and debug a Lens?
The CLAD skills cover Preview inspection, log analysis, the JS debugger, perf traces, and LEAF integration tests.

Add features

Can I connect an interactive voice agent, like a Grok bot?
Yes. Use built-in ASR (speech-to-text) plus a remote LLM through the Remote Service Gateway (supported OpenAI and Gemini realtime APIs), or your own backend over the Internet and WebSocket APIs. These services come for free. Third-party bots like Grok work only through your own backend. There's no native integration. You can use the fetch API.

Can I build smart-home gadgets?
Yes. Call your hub's API through the Internet, WebSocket, BLE, or Bluetooth APIs, or through your own backend.

Performance and safety

What are the performance bottlenecks, and how do occlusion, thermals, and battery work?

  • Frame budget: Overdraw, transparent layers, heavy shaders, and script work in UpdateEvent cost the most.

  • Occlusion: Use the world mesh and depth.

  • Thermals and battery: Profile with the Lens Studio performance tools (perf traces) and design for short, focused sessions.

  • Published specs: specs.com

How do I design for user safety and attention?
Keep users aware of their surroundings: don't block their central view, avoid fast or full-screen visuals, and design short sessions with clear exits. Platform guidelines cover comfort and safety, and Lens review enforces them.

Platform and roadmap

What are the main advantages of SPECS?
SPECS are standalone, see-through AR glasses with no phone tether. They come with hand tracking and a full developer platform (Lens Studio plus AI agents).

How do I carry a project forward to the 2027 glasses?
Build on the supported Lens Studio APIs and packages so your project carries forward. Snap will share migration guidance with releases.

Will SPECS support cross-platform reuse or previewing on other devices?
We understand how important this is, but it's not a priority right now. Porting has become a lot easier with AI.

As AI becomes more common, will AR matter for work and study?
Yes, it's where Snap is investing. Glasses give AI shared context (what you see and hear) and hands-free output. For example, a tool that helps people and AI inspect audio and visuals together, with provenance intact, fits well. Try it as an MCP connected to a SPECS Lens through a backend.

Still have a question? Ask it in the discussion forum. For setup help, join the Devpost Discord and ask in #specs-general.