mcpField
T
tailwind-context-resolver-mcp

tailwind-context-resolver-mcp

Local
github.com/vola-treblaDeveloper Tools0Updated 23w ago

About tailwind-context-resolver-mcp

tailwind-context-resolver-mcp is an MCP server for AI agents working with Tailwind CSS v3 projects. It reads the local configuration to expose design tokens and check utility classes against the resolved theme.

Key features

  1. 01Query resolved theme tokens
  2. 02Validate utility class strings
  3. 03Detect conflicting utilities
  4. 04Summarize configuration and plugins

Use cases

  1. 01Find available project colors and spacing values
  2. 02Check classes before adding a component
  3. 03Identify conflicting layout utilities

MCP support

It exposes tools for querying a project's resolved Tailwind theme, validating classes, detecting conflicts, and summarizing configuration to MCP clients.

Configuration

Local server config is the same for every client. Copy it into your client’s MCP config file.

mcp.json · Universal config
{
"mcpServers": {
"vola-trebla-tailwind-context-resolver": {
"command": "npx",
"args": [
"-y",
"tailwind-context-resolver-mcp"
]
}
}
}
Run commandnpx -y tailwind-context-resolver-mcp

Local servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.

Packages
npmtailwind-context-resolver-mcp0.2.0

Tools

7

Saved tool list · updated 3w ago

ParameterTypeRequiredDescription
filterstringNoOptional substring filter on token names
namespacestringYesDot-path into the theme (e.g. 'colors', 'colors.brand', 'spacing')
config_pathstringYesAbsolute path to tailwind.config.js or tailwind.config.ts

Details

From the project README

tailwind-context-resolver-mcp 🎨🐸

npm version npm downloads CI License: MIT

An MCP server that loads your project's tailwind.config.ts/js and exposes its actual design system to AI agents — so they stop hallucinating class names.


🤔 The Problem

AI agents generate Tailwind classes based on training data — the default Tailwind docs. Your project is not the default docs.

You have a custom color palette. A non-standard spacing scale. Maybe a prefix like tw-. Brand tokens like bg-brand-primary. The agent doesn't know any of this. It guesses.

The result:

// Agent confidently generates this:
<div className="bg-primary-500 text-brand p-18 tw-flex-center">

// Your project has:
// - bg-brand-primary (not bg-primary-500)
// - no "text-brand" token
// - spacing.18 = 4.5rem (ok actually)
// - no "flex-center" utility
// - no "tw-" prefix

The agent can't validate what it writes because it has no access to your resolved config. It's working from memory of the default theme — not yours.


✅ The Fix

This MCP server runs the Tailwind resolver locally and gives agents a typed, queryable interface to your actual config. Before writing a component, the agent can ask:

  • "What brand colors exist in this project?"
  • "Is p-18 a valid spacing value here?"
  • "Does this project use a custom prefix?"
  • "Is bg-brand-primary flex grid a valid class string?"

🛠️ Tools

resolve_theme_tokens

Query any namespace in the resolved Tailwind theme. Returns all design tokens as flat key-value pairs.

namespace: "colors.brand" → { primary: "#3b82f6", secondary: "#8b5cf6", danger: "#ef4444" }
namespace: "spacing"      → { "1": "0.25rem", "2": "0.5rem", "18": "4.5rem", ... }
namespace: "fontFamily"   → { sans: ["Inter", "sans-serif"], mono: [...] }

Use before generating components to discover what tokens actually exist.

validate_class_string

Validates a Tailwind className string against the project's resolved config. Returns valid classes, invalid (hallucinated) classes, and conflict warnings.

{
  "valid_classes": ["bg-brand-primary", "text-white", "p-4", "hover:bg-brand-secondary", "flex"],
  "invalid_classes": ["bg-fake-token", "text-brand"],
  "warnings": ["Conflicting multiple layout models: flex, grid"],
  "config_prefix": ""
}

Use to catch hallucinated design tokens before writing code.

detect_css_conflicts

Detects conflicting Tailwind utilities — e.g. flex + grid, or absolute + fixed on the same element.

{
  "conflicts": [{ "classes": ["flex", "grid"], "reason": "multiple layout models" }],
  "has_conflicts": true
}
get_config_summary

Returns a compact overview: Tailwind version, prefix, which theme sections are customized, active plugins.

{
  "tailwind_version": "3.4.19",
  "prefix": "",
  "theme_extensions": ["colors", "spacing", "fontFamily"],
  "total_colors": 142,
  "total_spacing": 34,
  "plugins": ["@tailwindcss/forms"]
}

Use first to understand the project's design system before querying specific tokens.


🚀 Setup

Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "tailwind-context-resolver": {
      "command": "npx",
      "args": ["-y", "tailwind-context-resolver-mcp"]
    }
  }
}
Cursor / VS Code / Any MCP client
{
  "tailwind-context-resolver": {
    "command": "npx",
    "args": ["-y", "tailwind-context-resolver-mcp"]
  }
}

📋 Requirements

  • Tailwind CSS v3 — v4 uses a CSS-based config format and is not supported (the server will tell you clearly)
  • Node.js 18+
  • A tailwind.config.js or tailwind.config.ts in your project

🔧 How It Works

The server uses the same config loading strategy as the Tailwind CLI:

  1. jiti loads your tailwind.config.ts at runtime — no ts-node required
  2. tailwindcss/resolveConfig merges your config with Tailwind defaults to produce the full resolved theme
  3. Tools perform token-based class validation — checking that bg-brand-primary maps to an actual colors.brand.primary token — without running the full PostCSS/JIT pipeline

This approach is fast, stable, and works with any Tailwind v3 project without additional configuration.


📖 Agent Workflow Example

1. get_config_summary       → understand the project's design system
2. resolve_theme_tokens     → query specific namespaces before writing classes
   (namespace: "colors.brand", "spacing")
3. validate_class_string    → validate the className string before committing it
4. detect_css_conflicts     → final sanity check for conflicting utilities

🐸 Part of the MCP Toolbelt

Built alongside:


License

MIT

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.