Getting Started

Last Updated: 2026-06-04

Prerequisites

Tool	Version	Notes
Node.js	≥18 (≥20.12 for the pipeline)	The pipeline scripts use --env-file-if-exists, which needs Node ≥20.12
Yarn	1.22.x (classic)	Package manager for this repo
Git	any recent	Required for the scraper's --clone mode

Setup

1. Clone the repository

git clone https://github.com/blendsdk/fluentui-mcp.git
cd fluentui-mcp

2. Install dependencies

yarn install

3. Build

yarn build

4. Run the tests (verify)

yarn build && yarn test

5. Run the server

# Serve the bundled v9 schema
yarn start:v9

# or the default (v9)
yarn start

The server speaks MCP over stdio. Startup diagnostics are printed to stderr; stdout is reserved for the MCP protocol.

Using the Server with an MCP Client

Point your MCP client (e.g. Claude Desktop, Cline) at the built binary:

{
  "command": "fluentui-mcp",
  "args": ["v9"]
}

For local development you can run it directly via node dist/index.js v9 after building.

Project Structure

data/v9/        Bundled schemas (raw + enhanced) — served at runtime
docs/v9/        FluentUI source documentation content (NOT this techdocs site)
techdocs/       This VitePress technical-architecture site
scripts/        Offline pipeline: scraper/ (ts-morph) + enhancer/ (LLM)
src/            Runtime: schema/, search/, formatters/, tools/, server.ts, index.ts
src/__tests__/  Vitest suites mirroring the source areas
.github/        CI/CD workflows
plans/          Planning artifacts (point-in-time; superseded by techdocs)

⚠️ Do not confuse docs/v9/ with techdocs/. docs/v9/ is FluentUI content payload; techdocs/ is this developer-facing architecture site.

Next Steps

• Read the System Overview to understand the pipeline-vs-runtime split.
• See the API Design for the 12 MCP tools.
• Review the Architecture Decisions for the "why".
• See the Development Workflow for the schema pipeline.