Skip to content
AI Builder

AI proposes. SynaptoDeck validates. You approve.

Other deck apps let AI fire your existing buttons. SynaptoDeck lets AI build new ones — and every proposal goes through your inbox before it touches your deck.

Authoring

AI calls propose_button with the button spec.

Validation

Schema check. Action whitelist. Risk classification.

Approval

You see every proposal. You ship it, edit it, or reject it.

Why this matters

Authoring isn't executing.

Stream Deck 7.4 shipped MCP support — but it's execute-only. Claude can fire existing buttons. SynaptoDeck's MCP authors them: AI proposes new buttons and pages, then you approve before they apply.

SynaptoDeck (authoring)
  • AI proposes buttons + pages.
  • You preview, edit, approve, or reject.
  • AI cannot fire buttons. No run_button tool by design.
  • Secrets never leave your machine. Substitution happens at fire time, not at proposal time.
Execute-only MCP
  • AI fires buttons that already exist.
  • No new buttons are authored from the chat side.
  • The approval surface is the deck itself, not an inbox.

Both models can coexist. The trade-off is where the human review happens — at run time, or at author time.

Safety invariants

The rules AI can't break.

Written into the MCP layer itself — not just policy.

No firing

AI cannot trigger an existing button. Only propose new ones.

No source-code access

AI sees deck config; never reads your codebase, files, or terminal history.

No secret values

Secrets are referenced by name. Substitution happens at fire time on your machine.

No bulk ops

AI proposes one button at a time. No mass-delete, no mass-replace.

Locked pages stay locked

Proposals targeting a locked page return locked:true; user approval gates everything.

Every action audited

Propose, approve, fire — all timestamped. Export to CSV from the Audit Log widget.

Works with

Bring your own client.

MCP-native everywhere. Direct API mode if you'd rather skip MCP.

Claude Desktop

MCP-native. The reference integration.

Claude Code

Build buttons from inside your terminal session.

Cursor

MCP server URL in settings; build buttons mid-edit.

Cline

VS Code agent with MCP tool support.

Don't want to wire up an MCP client? The AI Builder also takes direct Anthropic / OpenAI / local Ollama keys, or paste-JSON imports from any AI chat.

In the manifest

AI Builder features.

AI Builder via MCP

Connect Claude (or any MCP client) and ask it to build buttons for you. Every proposal lands in an inbox — you approve, SynaptoDeck applies. AI never writes to your deck directly.

Shipped in v0.43.0

AI Builder direct (Claude/OpenAI/Ollama)

Describe the deck you want in plain English. Pick from 3-8 suggestions; approve to apply.

Shipped in v0.41.0

Paste-JSON import

Don't want to wire up an API key? Paste JSON from ChatGPT / Claude / Gemini and we'll validate it.

Shipped in v0.38.0

AI Suggestion Inbox

AI suggests. SynaptoDeck validates. You approve. SynaptoDeck applies. Every change is reviewable, every action auditable.

Shipped in v0.38.0

Coming soon

AI Routines v1

Coming soon: AI Routines turn your deck into an agent. Set a goal, AI runs the steps, you keep the button as a one-tap repeat.

Target: v0.87.0+