Skip to content

Quickstart

The problem: a great answer can still fail if it uses the wrong docs syntax for the target framework.

mcp-zen-of-docs closes that gap by detecting the docs stack first, then resolving the native syntax the page actually needs.

A calm quickstart illustration showing a path with milestones toward a sunrise.

1. Install

# Recommended zero-install MCP runtime
uvx --from mcp-zen-of-docs mcp-zen-of-docs-server

# Or run the published container
docker run --rm -i ghcr.io/anselmoo/mcp-zen-of-docs:latest

If you want the CLI instead of the MCP server:

uvx --from mcp-zen-of-docs mcp-zen-of-docs --help

Quick installers for VS Code

Method VS Code VS Code Insiders
UVX Install Install
Docker Install Install

Naming guide

  • Package name: mcp-zen-of-docs
  • CLI command: mcp-zen-of-docs
  • MCP server command: mcp-zen-of-docs-server
  • MCP client server key: zen-of-docs

2. Connect your AI client

GitHub Copilot in VS Code

Add this to .vscode/mcp.json:

{
  "servers": {
    "zen-of-docs": {
      "command": "uvx",
      "args": ["--from", "mcp-zen-of-docs", "mcp-zen-of-docs-server"]
    }
  }
}

GitHub Copilot CLI

Launch copilot in your docs repository, run /mcp, and register a server named zen-of-docs with:

  • command: uvx
  • args: --from mcp-zen-of-docs mcp-zen-of-docs-server

Claude Desktop

Add this to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "zen-of-docs": {
      "command": "uvx",
      "args": ["--from", "mcp-zen-of-docs", "mcp-zen-of-docs-server"]
    }
  }
}

Restart Claude Desktop after saving the file.

Cursor

Add this to .cursor/mcp.json:

{
  "mcpServers": {
    "zen-of-docs": {
      "command": "uvx",
      "args": ["--from", "mcp-zen-of-docs", "mcp-zen-of-docs-server"]
    }
  }
}

3. Run the first real workflow

The most reliable pattern is:

detect → profile → act

Start with a concrete request such as:

Add a prerequisites note to the quickstart page.

A framework-aware assistant should then:

  1. Call detect to identify the docs stack.
  2. Call profile to resolve the primitive you need.
  3. Use that result when it writes or updates the page.

If you want the full explanation of that pattern, read Detect → Profile → Act.


4. What good output looks like

A correct answer is not just well written. It also renders correctly in the target docs stack.

  • Zensical should receive Zensical-compatible syntax.
  • Docusaurus should receive Docusaurus-compatible syntax.
  • VitePress and Starlight should receive their own native forms and caveats.

That is what mcp-zen-of-docs is designed to provide: source-truth context before content generation.


First questions

What if my config file lives in a subdirectory? Pass project_root explicitly to detect, for example project_root=./docs-site.

Can I use this with GitHub Copilot? Yes. Use .vscode/mcp.json for Copilot in VS Code, or /mcp in Copilot CLI, then use the copilot tool to generate instruction assets.

What should I read after setup? Start with the tool references for detect and profile, then continue to Troubleshooting if the target project is unusual.


  • Tools overview

    See all 10 tools and how they fit together.

    Browse tools

  • Detect → Profile → Act

    Learn the operating model behind the workflow.

    Read the guide

  • Troubleshooting

    Handle unusual layouts, detection misses, and rendering issues.

    Read troubleshooting