shadcn/ui MCP Server
shadcn/ui works by copying component source into your project rather than installing a dependency. That creates a problem: the AI doesn't know what your copy of a component looks like, so it routinely invents props that don't exist.
This server feeds the AI the actual shadcn/ui component source, demos, and dependency graph. Ask for "a data table with filtering" and it works from the real Table and Select definitions, so the generated code actually runs.
Install / Connect
# No npm package — install from source.
# See the repo's README for the exact command:
# https://github.com/Jpisnice/shadcn-ui-mcp-serverFollow the official README; env and args vary per server.
Why use it
Good for: frontend developers on shadcn/ui who've been burned by hallucinated props.
Before connecting: it calls the GitHub API to fetch component source, so without a token you'll hit the 60-requests/hour limit. Configure a read-only token.
Similar / alternative servers
todoist-mcp
A set of tools to connect to AI agents, to allow them to use Todoist on a user's
ai.imboard/dossier
MCP server for dossier automation standard - enables LLMs to discover, verify, a
ai.smithery/keremurat-json
Compare two JSON files deeply without worrying about key or array order. Detect
Frequently Asked Questions
Does it support shadcn/ui v4?
It fetches component source across versions; check the repo README for exact coverage.
How is this better than reading the docs?
It provides source-level detail — real props, internal implementation, which other components are used — which is more complete than doc examples and directly consumable by the AI.