@nomiedge/ai
v0.1.1
Published
An MCP server and static context exporter that lets an AI assemble correct Nomiedge Design Language surfaces — real component props, real tokens, real patterns, validated by construction.
Readme
@nomiedge/ai
An MCP server (and a static non-MCP context export) that serves the real Nomiedge Design Language registry, tokens, and patterns to AI tooling — so an agent can generate NDL code that's correct by construction instead of guessing at props and colors.
Run the server
npx @nomiedge/ai
# or, once installed:
nomiedge-ai serveMCP client config
{
"mcpServers": {
"nomiedge": { "command": "npx", "args": ["-p", "@nomiedge/ai", "nomiedge-ai", "serve"] },
},
}What it serves
- Resources (read-only, no params):
nomiedge://registry,nomiedge://tokens,nomiedge://patterns,nomiedge://rules. - Tools:
getComponent(name),listPatterns(),getTokens(group?),validateUsage(snippet).
The boundary
This package is a knowledge source, not an actor: no filesystem writes, no code execution beyond
static syntax parsing, no network calls. It never touches a consumer's project — that's the job of
nomiedge init/add (@nomiedge/cli), run by a human. See docs/adr/0006-mcp-read-only-knowledge-source.md.
Non-MCP fallback
import { buildContext } from "@nomiedge/ai";
const context = buildContext(); // { registry, tokens, patterns, rules, markdown }