Skip to main content
Glama
Carnegie Mellon Web Kilt MCP Server

Carnegie Mellon Web Kilt MCP Server

Official
by cmu-web

Carnegie Mellon Web Kilt MCP Server

An MCP server for the Carnegie Mellon Web Kilt, so an AI coding agent can set up the kit and build pages with it the way the docs describe.

It reads each kit release's own data (docs/kit.json and the stylesheets) from jsDelivr, so it knows every release from 0.3.0 on and never falls behind the docs. It only reads public docs: no sign-in, nothing stored.

Add It to Your Agent

The hosted server is at https://webkilt.andrew.cmu.edu/mcp (Streamable HTTP).

Claude Code:

claude mcp add --transport http cmu-web-kilt https://webkilt.andrew.cmu.edu/mcp

VS Code (.vscode/mcp.json):

{
  "servers": {
    "cmu-web-kilt": { "type": "http", "url": "https://webkilt.andrew.cmu.edu/mcp" }
  }
}

Cursor and most other agents take the same address in their MCP settings.

To run it on your own computer instead, over stdio:

npx -y cmu-web-kilt-mcp

Related MCP server: melta-ds-mcp

Tools

Tool

What It Does

search_components

Lists components that match a search, or all of them, with labels and docs links

get_component

One component's description, requirements, classes and example markup

get_install

The code to add the kit for a setup: Cascade, another CMS, a web app, Django, Node or a prototype

get_conventions

Class names, AP style and Title Case, headings, buttons and links, messages, colors, images and accessibility

check_html

Checks a page or fragment: unknown cmu- classes or data-cmu- hooks and, for a whole page, missing files, the cmu-app body class and the loading order

Each tool takes an optional version, such as 0.4.0; it defaults to the latest release. The server also has a set-up-kit prompt.

Develop

npm install
npm run dev          # Streamable HTTP on http://localhost:3000/mcp
npm run check        # lint, format, types, tests, build
node dist/cli.js     # stdio, after npm run build
node dist/cli.js --http 3333

The tests answer from a copy of kit 0.4.0's files in test/fixtures/, so they don't use the network.

See DEPLOYMENT.md for hosting.

License

BSD 3-Clause; see LICENSE.

Available Tools

5 tools
check_htmlCheck HTMLA
Read-only

Checks a page or a fragment for kit mistakes: cmu- classes or data-cmu- hooks the kit doesn't have and, for a whole page, missing files, the cmu-app body class and the loading order.

Inputs: html (required), version.

get_componentGet a ComponentA
Read-only

One component's full description, what it requires (its labels), its classes and example markup to copy.

Inputs: id (required), version.

get_conventionsGet ConventionsC
Read-only

The kit's rules for class names, writing (AP style, Title Case, names), headings, links and buttons, messages, colors, images and accessibility.

Inputs: topic, version.

get_installGet Install StepsB
Read-only

The code to add the kit to a project, for one kind of setup, with the files and their loading order.

Inputs: setup (required), version.

search_componentsSearch ComponentsA
Read-only

Lists the kit's components (and the template's, labeled Template) that match a search, with their labels and docs links. With no query, lists them all.

Inputs: label, query, version.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.6/5.0

    Scored across 5 tools

    Disambiguation5/5

    Each tool targets a distinct action on a distinct object: search_components lists matches, get_component retrieves one component's details, get_install returns setup code, get_conventions returns rules, and check_html validates markup. There is no meaningful overlap between listing, fetching, and validating.

    Naming Consistency5/5

    All five tools use a consistent snake_case verb_noun pattern (search_, get_, check_). The verbs accurately reflect the action, and there is no mixing of conventions.

    Tool Count4/5

    Five tools is on the low end but covers the discover/read/install/validate lifecycle without redundancy. Slightly more surface (e.g., a separate token or template lookup) could exist, but nothing feels extraneous.

    Completeness4/5

    The surface covers finding components, reading their docs, installing the kit, retrieving conventions, and validating HTML—the core workflows for building a page with this kit. Minor gaps remain, such as retrieving full page templates or searching within the conventions, but agents can work around these.

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      A
      quality
      B
      maintenance
      It lets AI coding agents look up design tokens, component specifications, and prohibition rules on demand, then validate generated HTML or JSX with the same lint logic used by CI and editor hooks so violations are caught and corrected immediately. It also exposes the design constitution and rule sets as resources for reference during generation.
      6
      272 npm
      202
      MIT
    • A
      license
      A
      quality
      B
      maintenance
      Enables AI coding agents to build UI with a design system's tokens, component inventory and prop signatures, page templates, and starter snippets, while tracking free/pro availability and live-updating design data.
      11
      102 npm
      2
      MIT