Vue Docs MCP

The Problem

LLMs have a knowledge cutoff. The Vue ecosystem evolves. When you ask your AI assistant about Vue:

• Code examples are outdated and based on old training data

• Hallucinated APIs that don't exist in the current version

• No way to cite or verify answers against the official docs

The Solution

Vue Docs MCP gives your AI assistant direct access to the Vue ecosystem documentation through the Model Context Protocol. Every answer is grounded in the official docs.

The hosted server at mcp.vue-mcp.org is free. No API keys, no setup required.

Supported Frameworks

Each framework gets its own tools, resources, and prompts. See all frameworks.

Quick Start

Claude Code

claude mcp add --scope user vue-docs --transport streamable-http https://mcp.vue-mcp.org/mcp

Cursor

Or add to .cursor/mcp.json:

{
  "mcpServers": {
    "vue-docs": {
      "url": "https://mcp.vue-mcp.org/mcp"
    }
  }
}

Other Clients

For Claude Desktop, Windsurf, VS Code, JetBrains, Zed, ChatGPT, and more, see the full client list.

Any MCP client that supports streamable HTTP works with:

https://mcp.vue-mcp.org/mcp

What's Included

Tools (per framework):

Resources (per framework):

Prompts (per framework):

Benchmarks

Evaluated by an LLM judge (Gemini, temperature 0) scoring retrieved documentation on relevance, completeness, correctness, API coverage, and conciseness (1-5 scale).

Each question has ground-truth answers with expected API names and documentation paths. An LLM judge scores the retrieved documentation context. API recall measures whether expected API names appear in the response. See the eval/ directory and framework pages for the full breakdown by difficulty, question type, and judge dimension.

Context7 is a general-purpose documentation server covering 9000+ libraries. Vue Docs MCP is purpose-built for the Vue ecosystem. The comparison shows the quality advantage of specialization, but Context7's breadth is a genuine strength for multi-ecosystem projects. Context7 also returns Vue 2 content for some Vue 3 questions, which affects its scores.

Run make eval-compare to reproduce these results.

How It Works

1. Structure-aware chunking. Docs are parsed respecting their heading hierarchy, keeping code examples paired with explanations.

2. Hybrid search. Every query runs dense semantic search (Jina embeddings) and BM25 keyword search simultaneously.

3. Smart entity detection. API names detected with typo tolerance, synonym lookup, and fuzzy matching.

4. Cross-reference expansion. Related documentation sections are automatically pulled in.

5. Reranking. Candidates reranked by Jina for precision.

6. Readable reconstruction. Results reassembled in documentation reading order.

No LLM is used at query time. See How It Works for the full technical breakdown.

Self-Hosting

git clone https://github.com/joelbarmettlerUZH/vue-mcp.git
cd vue-mcp
make bootstrap          # Clone Vue docs + install deps
cp .env.example .env    # Add your API keys
make ingest             # Index the documentation
make serve              # Start the MCP server

See the Self-Hosting guide for Docker Compose deployment and configuration details.

Development

make help         # Show all available commands
make test         # Run tests (skips integration tests)
make pr-ready     # Lint + format + test (run before committing)

Documentation

Full documentation at vue-mcp.org:

• Getting Started: Connect in 30 seconds

• Supported Frameworks: Tools, benchmarks, and roadmap per framework

• MCP Clients: Setup guides for 10+ clients

• How It Works: Ingestion and query pipeline internals

• Self-Hosting: Run your own instance

License

FSL-1.1-ALv2. Functional Source License, Version 1.1, with Apache License 2.0 future grant. Free for internal use, education, and research. Converts to Apache 2.0 two years after each release.