MCP server
manniche-ui
From idea to component.
Give your coding agent access to Manniche UI. It finds the right component, reads the code and the example, and installs it in your project.
Claude Code
claude mcp add --transport http manniche-ui https://mikkelmanniche.dk/api/mcpCodex
codex mcp add manniche-ui --url https://mikkelmanniche.dk/api/mcpCursor
{
"mcpServers": {
"manniche-ui": {
"url": "https://mikkelmanniche.dk/api/mcp"
}
}
}Run the command in your terminal and start a new agent session. Cursor reads the JSON from .cursor/mcp.json.
The address for other MCP clients
https://mikkelmanniche.dk/api/mcpTry an example
Pick a question and watch what happens: the question, the call to the server and the answer. The answers are real responses from the server, recorded in advance.
Find a component
Does Manniche UI have a chat input for my agent?
search_components{ "query": "chat input" }Running …DoneThe server’s response (JSON, shortened)
{
"content": [
{
"type": "text",
"text": "- prompt-input (agent): A chat input that grows with the text, takes files, and turns into a stop button while the agent writes."
}
],
"isError": false
}Yes, Prompt input fits. Shall I install it?
prompt-inputAI interfacesA chat input that grows with the text, takes files, and turns into a stop button while the agent writes. With `halo`, a band of primary light runs round the border while there is something to send. With `variant="glass"` it is a frosted panel for photos, and with `collapsed` it rests as a small pill until focused.
Install two
Install prompt-input and tool-approval in my app.
get_install_plan{ "names": ["prompt-input", "tool-approval"] }Running …DoneThe server’s response (JSON, shortened)
{
"content": [
{
"type": "text",
"text": "Install plan for: prompt-input, tool-approval …"
}
],
"isError": false
}The plan has four steps. I’ll run them in your project:
- Only if the project has no components.json
npx shadcn@latest init - Add both components
npx shadcn@latest add @manniche/prompt-input @manniche/tool-approval - Import them from components/
import { PromptInput } from '@/components/prompt-input' - Check them in the running app
Read the code
How do I use hold-to-confirm?
get_component{ "name": "hold-to-confirm" }Running …DoneThe server’s response (JSON, shortened)
{
"content": [
{
"type": "text",
"text": "# Hold to confirm (`hold-to-confirm`) …"
}
],
"isError": false
}The button fills while it is held and calls onConfirm when it is full:
import { HoldToConfirm } from '@/components/hold-to-confirm'
<HoldToConfirm onConfirm={() => console.log('store closed')} doneLabel="Store closed">
Hold to close the store
</HoldToConfirm>
Dependencies: lucide-react, utils
The server only reads the library. Your agent installs locally and should check that the component works in your app.
Tools 4
list_componentsEvery component with its category and a line about each.
search_componentsSearches names, descriptions and categories.
get_componentDescription, dependencies, install command, example and the full source.
get_install_planThe steps to install one or more components in a project.
Or use shadcn’s own MCP
If your project already has components.json, shadcn’s local MCP server can find the components too once the registry is added.
{
"registries": {
"@manniche": "https://mikkelmanniche.dk/lab/r/{name}.json"
}
}npx shadcn@latest mcp init --client claude