This version was released on June 18, 2026.
A new kind of consumer
There’s a shift happening in how design systems are consumed. Design systems were built for people to read, and people still do that. But they’re no longer the primary audience.
Because, lately, there’s a second consumer. The AI tools your team builds with reach into the design system constantly, and on most teams that happens far more often than a person ever opens the docs.
The common approach so far is to hand an AI tool the whole design system through one MCP and let it sort things out. That somewhat works. But the AI gets a lot of information it doesn’t need, and it can’t always tell the difference. The output drifts toward things that don’t matter, and you pay for the excess in quality and AI tokens. The bigger the system, the more teams pulling from it, the more that adds up.
This release is about control and governance: giving each team, tool, and agent the right part of your design system — and nothing more. No more handing over everything and hoping the AI can figure out what’s important.
Contexts
Put simply, a Context is a configurable container. You connect your design system, and then use dynamic filters to determine exactly what the context contains: from the whole system down to specific pages, components, or design tokens. You can also build a context that contains organizational knowledge (files and skills), with no design system attached.
Each context publishes as its own MCP server, so anything that connects sees only what you put in. One design system can feed many different contexts, each shaped for its unique consumer.
For example:
- A product team gets most of the system: code components, their APIs, and the guidelines for using them. While a marketing team gets a smaller slice, focused on brand, tokens, and content guidelines.
- Designers get Figma components and usage guidelines.
- Engineers get detailed APIs and usage examples for the React components.
- An autonomous agent checking every pull request against content guidelines gets only the content rules, without the noise of the rest of the system.
It’s a deterministic layer over a non-deterministic one. What the AI can consume is pre-defined — not left for the model to sort out at runtime.
Filter only the design system data that you want your AI agents to work with.
Compose contexts together
Once you’re running more than a handful of contexts, you’ll notice how much they have in common. Most of them share the same brand foundation, the same accessibility guidelines, the same core setup — configured manually each time.
With context composition, you can build a context from shared, modular pieces: inherit a common foundation and layer the context-specific parts on top. When the shared context changes, everything built on top of it updates too, so a team running dozens of contexts maintains one source instead of dozens of copies.
This is what makes the multi-team, multi-brand case actually manageable.
Scoped MCP servers
Our MCP server(s) got a huge upgrade in this release, shaped by feedback that you shared with us (thank you!).
Supernova MCP now has a semantic search tool covering your documentation, files, and skills. Your AI agents can now find things by meaning instead of browsing through the whole docs and trying to guess where things live.
Additionally, the MCP now knows your code components: their props, anatomy, and usage examples. Ask how a component actually works and the answer is readily available, backed by your real design system.
But the biggest change is tied to our new Contexts feature. Each MCP serves only what you have scoped. No more dumping the whole design system on the agent and hoping for the best.
This leads to:
- less noise in MCP responses
- generally better answers
- fewer tokens spent on things the task never needed
We've also made significant performance improvements, so responses come back quicker.
Your organizational knowledge: Files and skills
Sometimes there is knowledge you want to give your AI agents that doesn’t really belong in the design system, for example: internal conventions, naming rules, a doc explaining how your team works. With Files, you can upload your markdown or PDFs, attach it to a context, and serve it through the MCP alongside the rest of your data.
Skills go a step further. They’re a new type of knowledge written specifically for AI: the guidelines, rules, and conventions you want an agent to follow directly. Attach your skill markdown files to a context to distribute them via MCP.
It's our first knowledge format made for AI rather than for people, and an early first version at that. Documentation has always been written for humans to read; Skills is a step toward knowledge built for agents instead.
Distribute contexts as ready-to-use plugins
Once a context is set up, getting it into your tools should take seconds. Now it does: export it as a plugin, drop it in the repo, and install in a few clicks. The MCP connection and Skills come bundled.
At launch, we support exporting contexts to every major AI provider format:
- Claude Code plugin
- Cursor plugin
- Codex plugin
- and a generic Skills export for everywhere else.
And because it all comes from one context, the guidance stays consistent wherever it lands, rather than drifting across the tools and repos it usually gets scattered between.
Feature preview: Context feedback
A context is only as good as what it contains, and the people using it are the first to notice when something seems off — now they can tell you directly.
Working from their AI tool, anyone can tell the agent to send feedback to the team that owns the context: a missing guideline, a wrong token, a gap they ran into. It goes through a tool on the MCP and is stored with the context, so the team can find all the feedback in one place and act on it.
We’re also experimenting with collecting feedback automatically, with the agent flagging friction as it works.
We’re shipping it as a preview to learn early. Give it a try and tell us what’s useful, what's missing, and where you’d want it to go. We’re genuinely building this one together.
A redesigned primary navigation
Supernova has grown a lot, and the navigation hadn't kept up. So we rebuilt it to better expose and structure everything the platform offers, making it clearer what's available and where to find it.
Moving around is faster too: jumping between sections and different design systems now takes fewer clicks and less hunting.
Availability
- Contexts are available on all plans.
- Dynamic filters are available on Pro and above.
- Context composition is available on Enterprise.
MCP consumers are part of every plan now. Free includes 5. All our current Pro and Enterprise customers will get up to 50 as part of their current plan; if you need more, please reach out to our sales team at sales@supernova.io.