Your agents build. Your judgement leads.
Uxi turns Claude Code into an experience-design partner.
Every output lands in one review hub, where you shape it and approve it before your agent takes the next step.
How it works
01 · Run a workflow. Start Discover, Define, Design system, Develop, Audit, or Deliver from the Uxi sidebar.
02 · Your agent drafts it. Moodboards, flows, design systems, wireframes, and screens are written as real HTML in your project's uxi-output folder.
03 · You shape it. The Uxi review hub opens in your browser. It shows the output at desktop and mobile size, beside the controls that change it: pick a direction, reorder a flow, mark where a person must decide, rewrite a headline, trim the scope.
04 · You approve. Your decision goes straight back to the agent. It continues only with what you approved, and asks before touching files outside that scope.
Agent proposes → You shape → You approve → Agent continues
Six workflows. Your judgement throughout.
|
Workflow |
What you review in the hub |
| 01 |
Discover · /uxi-discover |
Research brief, competitor analysis, research plan, and the user journey as it is today |
| 02 |
Define · /uxi-define |
Problem statement, editable user flows (including agent and human-review steps), wireframes for every screen |
| 03 |
Design system · /uxi-branding |
Moodboard with real references, three visual directions, logo concepts, brand guidelines, design system, tokens.css, landing page |
| 04 |
Develop · /uxi-develop |
High-fidelity screens with every state, a screen overview, a clickable prototype of the approved flow |
| 05 |
Audit · /uxi-audit |
Usability, accessibility, and consistency findings next to the UI they're about; you choose what to fix |
| 06 |
Deliver · /uxi-deliver |
The implementation plan, then each built screen side by side with the approved design |
The hub keeps every stage together, so the project reads as one story: what was proposed, what you decided, and what the agent did next. A saved copy is written to uxi-output/index.html at the end of each workflow.
Figma, for what it's good at
Uxi registers the official Figma MCP server and uses it to bring things in: references for a moodboard, an existing design to audit, or screens to build from. Your outputs live in your project as HTML, so you review them in one place and your agent builds from exactly what you approved. Figma export only happens if you ask for it.
Get started
- Install Claude Code.
- Install Uxi, open a project folder, and click the uxi. icon in the activity bar.
- Sign in, then click Set up Uxi. This installs the workflows, the review hub, and the Figma MCP for every project, and optionally stores a Gemini or OpenAI key for image generation (kept in VS Code's secret storage).
- Click any workflow. A terminal starts Claude with it, and the review hub opens with the first output.
Open review hub in the sidebar reopens the live hub while an agent is running, or the saved uxi-output/index.html when it isn't.
What stays on your machine
- The review hub is served from your computer (127.0.0.1) by the agent's Uxi MCP server. It accepts decisions only from its own page, and generated pages preview in a sandbox.
- Session records live in
.uxi/ in your project. They hold summaries, decisions, and file paths, never raw prompts, secrets, or file contents. Add .uxi/ to .gitignore.
- Uxi records observable actions and decisions. It does not expose or store the model's private reasoning.
Also for Claude Code and Codex, without VS Code
The extension's plugin folder is also a Claude Code and Codex plugin. After installing, it lives inside the extension folder:
claude --plugin-dir ~/.vscode/extensions/uxicraft.uxi-0.4.1/plugin
uxi. Human judgement. Agent momentum. · uxicraft.com
