使用说明

Slice.js Documentation MCP

An MCP (Model Context Protocol) server that provides access to Slice.js documentation from the official GitHub repository. This server allows AI assistants and tools to query, search, and retrieve documentation seamlessly.

Features

Dynamic Documentation Discovery: Automatically discovers and indexes all documentation files from the GitHub repo
Intelligent Caching: Infinite session cache to minimize API requests and improve performance
Full-Text Search: Search across all documentation with keyword matching
Content Retrieval: Fetch specific documentation pages or the complete documentation bundle
Lazy Initialization: Docs structure is loaded on-demand when first needed

Installation

Global Installation (Recommended)

npm install -g slicejs-mcp

Using npx (No Installation Required)

npx slicejs-mcp

Usage

The MCP server runs as a stdio-based service, perfect for integration with AI assistants and MCP-compatible tools.

Basic Usage

npx slicejs-mcp

Integration with MCP Clients

This server is designed to work with MCP-compatible clients. When launched, it exposes the original framework tools plus namespace-aware tools for the Slice.js Components catalog:

Tools

1. `list_docs`

Returns a list of all available documentation sections and categories.

Parameters: None

Response: JSON array of documentation items with id, title, and path.

Example:

[
  {
    "id": "getting-started",
    "title": "Getting Started",
    "path": "markdown/getting-started.md"
  }
]

2. `search_docs`

Searches across all documentation using keywords or phrases.

Parameters:

query (string, required): Search term
max_results (number, optional, default: 5): Maximum number of results

Response: JSON array of search results with snippets and metadata.

3. `get_doc_content`

Fetches the full content of specific documentation page(s).

Parameters:

doc_id (string or string[], required): Documentation ID(s) to fetch
include_metadata (boolean, optional, default: false): Include additional metadata

Response: JSON object(s) with document content, title, and optional metadata.

4. `get_llm_full_context`

Fetches the complete documentation bundle (~2000 lines) for comprehensive LLM context.

Parameters: None

Response: Complete documentation text

Note: This consumes considerable tokens but provides all documentation in one request.

Framework catalog tools

These tools make the documentation source explicit while preserving the original tools above:

slice_framework_list_docs
slice_framework_search_docs
slice_framework_get_doc

They read from slicejs_docs and return authored Markdown plus provenance metadata.

Slice.js Components catalog tools

These tools provide access to Visual and Service component documentation:

slice_components_list_docs
slice_components_search_docs
slice_components_get_doc

The Components tools read the catalog manifest published by slicejs-components and retrieve the original Markdown document, including its examples and :::script blocks. They do not execute documentation examples or modify the project.

The catalog manifest is expected at:

https://raw.githubusercontent.com/VKneider/slicejs-components/master/agent/docs-manifest.json

Examples

List all documentation

// Via MCP client
await callTool("list_docs", {});

Search for routing information

await callTool("search_docs", {
  query: "routing",
  max_results: 3
});

Get specific documentation

await callTool("get_doc_content", {
  doc_id: "getting-started/routing"
});

Get full documentation context

await callTool("get_llm_full_context", {});

Architecture

Framework source: Documentation fetched from https://github.com/VKneider/slicejs_docs
Components source: Markdown and manifest fetched from https://github.com/VKneider/slicejs-components
Caching: Infinite session cache prevents redundant API calls
Initialization: Lazy loading of document structure on first tool use
Rate Limiting: Optimized to stay within GitHub API limits (60 req/hour)

The two catalogs are separate because framework behavior and component APIs have different versioning and ownership. The MCP uses the same list/search/get pattern for both sources, while the authored Markdown remains the canonical content.

Environment overrides are available for local development and mirrors:

SLICE_FRAMEWORK_DOCS_BASE_URL
SLICE_FRAMEWORK_DOCS_REVISION
SLICE_FRAMEWORK_DOCS_API_BASE
SLICE_COMPONENTS_MANIFEST_URL
SLICE_COMPONENTS_RAW_BASE_URL

Development

Prerequisites

Node.js >= 18
npm or yarn

Setup

git clone <repo>
cd slicejs-mcp
npm install
npm run build

Running Locally

npm start
# or
node dist/index.js

Testing with MCP Inspector

npx @modelcontextprotocol/inspector node dist/index.js

Contributing

Contributions welcome! Please ensure:

All tools maintain backward compatibility
Cache behavior is preserved
Error handling is robust

License

MIT