Skip to content

shadcn/ui in your BoltAI chats

Add one entry to BoltAI's MCP config and any model with tool use can search components and blocks and hand you the install. You'll need shadcn.io Pro on top of your BoltAI license.

Kind
App
Goes in
Plugins → MCP server config
Connects
Through mcp-remote
Team config
No

Add it to BoltAI's mcp.json

BoltAI runs local MCP servers from one JSON file, and this entry is one of them. Signed in with Pro, the config beside the steps carries your own URL.

  1. Copy the entry

    With Pro, the config holds your MCP URL, token included, as the last of mcp-remote's arguments. No token yet? Create one on your account.

  2. Open Settings → MCP Servers

    In BoltAI 2, add the shadcnio entry inside mcpServers in ~/.boltai/mcp.json, next to any servers already there. BoltAI 1 edits the same file from Settings → Plugins and its plus button.

  3. Start or reload shadcnio

    The first start fetches mcp-remote through npx, so give it a moment. Once it runs, its 22 tools show as available on the same screen.

  4. Switch its tools on in a chat

    In a new chat, check Tool Use is on, open Tools in the composer and enable shadcnio: all 22 tools, or only the ones the task needs.

Plugins → MCP server config
Sign in to get your config
Sign in with a Pro account, and this holds your MCP URL, ready to paste.

How it reads, with YOUR_TOKEN where your token goes:

{
  "mcpServers": {
    "shadcnio": {
      "args": [
        "-y",
        "mcp-remote",
        "https://www.shadcn.io/api/mcp?token=YOUR_TOKEN"
      ],
      "command": "npx"
    }
  }
}

~/.boltai/mcp.json is local-only: BoltAI doesn't sync it to BoltAI Cloud or your other devices, so your token stays on this Mac. If it leaks, regenerate it.

Where the server goes, by app
AppWhere it goesThen
BoltAI 2 for MacSettings → MCP Servers, saved in ~/.boltai/mcp.jsonStart or reload it, then Tools in the composer
BoltAI 1 (1.34 or later)Settings → Plugins, the plus button, Save ChangesEnable it in a chat's plugin dropdown
BoltAI on iPhone and iPadRemote servers only: iOS can't start mcp-remoteUse it from the Mac

See it running

Settings → MCP Servers lists each local server with its state: a running shadcnio shows its tools, and one that didn't come up shows a status such as Startup timeout instead.

  1. Open Settings → MCP Servers and find shadcnio running, its 22 tools listed.
  2. Start a chat with a model that supports tool use, open Tools in the composer and switch shadcnio on.
  3. Send the prompt below: the model calls whoami, and the reply names your account and your remaining calls for the minute.
Prompt
use shadcnio and tell me which account this chat is signed in with

Three chats, with the MCP and without

Two plain chats and one with a linked project folder: the prompt, every call, and what you get back. Flip the MCP off to see the same prompt answered from memory.

Your agent
  1. shadcn.io MCP connected · 22 tools
  2. You: Plan a launch page for my Mac menu-bar app, with a waitlist.
  3. MCP call: plan_page({ brief: "launch page for a Mac menu-bar app with a waitlist", framework: "react", kind: "marketing" })Returns: Hero, features, FAQ and CTA, each with two alternatives
  4. MCP call: get_block({ names: ["hero-centered-email-capture", "features-bento-grid", "faq-accordion", "cta-centered-dark-email-capture"], include: ["install"] })Returns: The four blocks' install, as one command to paste
  5. Agent step: Wrote the section order and the command into the replyReturns: Four blocks, top to bottom, ready for Terminal
  6. Your agent: Here's the page from top to bottom, with one command that adds all four blocks; run it in your project, then swap in your own copy.

In your project

Launch page

  • Four sections, in the order the plan picked
  • One command to paste in your project's Terminal

1 prompt → 2 MCP calls

Prompts for a BoltAI chat

Copy one into a chat with shadcnio's tools on. Naming the server steers the model to it; with a linked folder, name the file too.

01
use shadcnio to plan a landing page for my podcast app and give me one command for every block
02
use shadcnio to compare three pricing blocks that switch between monthly and annual billing
03
use shadcnio to write a Vue 3 combobox for choosing a country, every file with its path
04
use shadcnio to find Phosphor icons for upload, share and archive, and give me the imports
05
use shadcnio to turn Stripe's DESIGN.md into CSS variables for light and dark mode
06
use shadcnio to show me the Slider's props, with an example for a price range
07
use shadcnio to add an onboarding stepper to the welcome screen in my linked project
08
use shadcnio to list what a new SvelteKit app needs before its first shadcn component

Make it work better in BoltAI

BoltAI's own controls, from tool picking to projects and agents, decide how the model uses shadcnio.

  • Pick tools, not the whole server

    BoltAI 2 lets you enable single tools from a server. A quick component question needs only search_components and get_component; a full page adds plan_page and get_block. Fewer tools means less for the model to read on every request.

  • Keep it to one project

    Open the project in the sidebar and choose Manage Plugins in its header to turn on shadcnio there and nowhere else. Give the project a default agent, and new chats start with its instructions, model and plugins.

  • Link the project folder

    Since 2.16, a project can link folders on your Mac (beta): the model reads package.json, edits files once you approve, and reads AGENTS.md or CLAUDE.md from the folder. One line there makes shadcnio the default.

    Before writing UI, search the shadcnio MCP for a component or block and use the install it returns.
  • An agent for UI work

    Build an agent whose instructions send UI questions to shadcnio, with its tools switched on. Since 2.18, a new chat's tools follow the agent you pick, so choosing it is the whole setup.

  • Point it at npx

    Local servers start from your shell's environment, and BoltAI 2 has no PATH field. If the server can't find npx, set command to its absolute path; MCP Runtime ENV under Settings → MCP Servers controls which variables get passed.

    $ which npx
    # put the path it prints in "command", in place of npx
  • Long answers and step limits

    A Vue or Svelte install plan can run long: BoltAI continues a reply that hits the output limit, up to three times (Settings → General). An agent run stops at its step limit, 30 by default, raised under Agent Loop Control in the chat's inspector.

When shadcnio won't answer

What BoltAI shows, and the fix. Its Console.app logs (opens in a new tab) have an MCP category, and the troubleshooting guide has the rest, for every client.

shadcnio shows Startup timeout in Settings → MCP Servers.
The first start downloads mcp-remote, which can take longer than BoltAI waits. Reload the server once it's done; on a version before 2.12, update, since 2.12 fixed servers timing out at launch.
The server never starts, and the log says npx can't be found.
Install Node.js, which brings npx, or set command to the path which npx prints. BoltAI starts local servers from your shell's environment.
shadcnio stopped connecting after you made a new token.
The URL in ~/.boltai/mcp.json still carries the old one, so the server answers 401. Copy the config from this page again, replace the entry and reload shadcnio.
The MCP log shows 403 when shadcnio starts.
Your token is valid, but the account has no active Pro plan. Renew it and reload the server: the entry itself doesn't change.
Tools in the composer doesn't offer shadcnio.
Check that it's running in Settings → MCP Servers and that Tool Use is on for the chat. A model without tool use, or a project or agent with its own tool list, can hide it too.
The model answers from memory and never calls a tool.
Its tools may be off in this chat: switch them on from Tools. Then name the server (use shadcnio…), or put a line in the linked folder's AGENTS.md.

Get the MCP for BoltAI with Pro

The MCP comes with every Pro plan, with every block and premium template, for you or your whole team. Monthly, yearly or lifetime.

4.9/5 out of 5 · 150+ reviewsfrom 150+ founders and developers.

  • Secure checkout via Stripe
  • Code lives in your repo
  • Keep what you've shipped if you cancel
  • License terms

Frequently asked questions

The shadcn MCP in BoltAI: setup, limits, and the rest.

4.9 out of 5 from 150+ developers

Bring the registry into BoltAI

6,231 blocks, 294k icons and every component, for whichever model you chat with in BoltAI, in every Pro plan.

Get Pro