Candor UI

Guides / Lovable

Candor UI with Lovable

Connect the docs MCP server as a chat connector, put the brief in the project's Instructions, and ask for a chat.

  1. 01Add the registry

    Lovable builds React apps with shadcn/ui, so ask it to add @candor to components.json and install from it with npx shadcn@latest add @candor/<name>. Lovable runs the install; you can check components.json in the code view.

    JSON · components.json
    {
  "registries": {
    "@candor": "https://candor.themethodic.com/r/{name}.json"
  }
}

    See:Get started

  2. 02Add the docs MCP server

    The docs MCP server at https://candor.themethodic.com/mcp is remote (streamable HTTP), read-only and needs no key: it searches these docs, reads any page and suggests components for a task. It serves docs, never paid source. Lovable takes a custom MCP server as a chat connector: open Connectors, choose Add custom then MCP server, name it, paste this address as the Server URL, choose No authentication and click Add server. A chat connector gives Lovable context while you build; it isn't part of the published app, which is what you want for docs. If MCP server is greyed out, a workspace admin has turned custom servers off.

    The server's address

    See:Lovable: custom MCP servers

  3. 03Give it the brief

    There's no skills directory to install into, so put the brief where the tool reads instructions: Project settings → Instructions (plain text, up to 10,000 characters), or the workspace's Instructions for every project. The llms.txt index lists every docs page as Markdown; llms-full.txt has every component with its props.

    Text · Brief
    Use Candor UI (shadcn/ui components for AI agents): https://candor.themethodic.com/llms.txt
Registry: add "@candor": "https://candor.themethodic.com/r/{name}.json" under "registries" in components.json, then npx shadcn@latest add @candor/<name>.
Start with thread, message and composer; tool-call for tool calls.
Rules: model output is text, never HTML or links; unknown values stay unknown; a control that can't be used says why.

    See:llms.txtllms-full.txt

  4. 04Try it

    With the server and the skill in place, the agent can look up Thread, Message, Composer and Tool call in these docs and install them with the CLI. Check what it installed against candor-lock.json.

    An example prompt

    See:ThreadComposerTool callPatterns