Skip to content
Index
Design + AI

Luia MCP

Knowledge infrastructure for AI-assisted design.

Context

Luia is an MCP server that gives AI design tools persistent memory. Install it in Claude and it searches your team's accumulated conventions before answering, then records new decisions back to the store as you make them. The result: an AI that knows how your team works, and keeps learning.

It started as a fix for my own workflow and grew into infrastructure. The premise is simple: the model is not the bottleneck anymore, the context is. A team that feeds its accumulated judgement into every session gets an assistant that designs like the team. A team that does not gets a stranger every morning.

TypeMCP Server
StackTypeScript
DurationOngoing
RoleDesigner + Builder
Integrations
Claude
Figma
Cursor
Codex
The cycle

Luia runs as a loop, not a library. Work happens, decisions get captured as they are made, and the next session starts already knowing what the last one learned. Two design choices carry it. Recall is automatic: the search tool is described so the AI reaches for it unprompted, because no designer remembers to ask for their own conventions. Capture is a single call with no ceremony, because a documentation step with friction is one that never happens.

01DesignDesign as usual. Luia runs silently.
02CaptureDecisions and patterns are extracted.
03SynthesizeNew knowledge merges into the store.
04ApplyNext session starts with full context.
Knowledge structure

Every design decision is captured as a node: a spacing rule, a naming convention, a component behavior. All attributed to the designer who made the call and versioned across sessions. Luia surfaces these decisions across every project in the structure so nothing gets lost between sessions or teams.

Luia graph showing how nodes inform each other
Each node shows what informed it and what it informs.

The graph shows how decisions relate to each other. Every node carries two directions: what informed it and what it goes on to inform. A spacing choice from one project becomes context for a layout decision in the next. The knowledge compounds across projects, not just within them.

The whole structure is deliberately boring: one Markdown file per decision, frontmatter for metadata, a folder for the lot. No database, no backend, nothing to host or maintain. It versions with git, and reads like something a human wrote, because most of it is. The format is the feature. Knowledge you cannot open in a text editor is knowledge you will eventually lose. Every rule carries its reasoning, which is also what lets the AI apply it to a case nobody anticipated.

The wiring
spacing.mdnaming.mdcomponents.md
Luia MCPsearch · read · record · list
Every sessionReads before answering, writes as decisions are made.
For teams

For an individual, Luia is memory. For a team, it is alignment. The same store that stops one designer from re-explaining conventions stops five designers from drifting apart. Because it is just a folder of Markdown, sharing it is git: each person points the server at the same repo, pulls to inherit everyone's decisions, and pushes their own. Conventions arrive as readable diffs you review like any other change, rather than as a document nobody opens.

Without LuiaWith Luia

Every AI session starts from zero. Conventions get re-explained, or forgotten.

Every session inherits the full team graph. Context is never re-entered.

A decision made on one project doesn't reach the next. Knowledge lives in heads.

A spacing rule from one project becomes context for the layout decision in the next.

A new designer gets up to speed through documentation that's always out of date.

A new designer's first session already knows how the team works.

Next projectSeeNLHigh-end motion for a film platform

All work