Shared Knowledge MCP Server

This is a knowledge base MCP server that can be used in common with various AI assistants (CLINE, Cursor, Windsurf, Claude Desktop). It utilizes Retrieval Augmented Generation (RAG) to realize efficient information search and utilization. By sharing the knowledge base between multiple AI assistant tools, it provides consistent information access.

Features

A common knowledge base can be used across multiple AI assistants
High-precision information retrieval using RAG
Type-safe implementation using TypeScript
Supports multiple vector stores (HNSWLib, Chroma, Pinecone, Milvus)
Extensibility through abstracted interfaces

install

git clone https://github.com/yourusername/shared-knowledge-mcp.git
cd shared-knowledge-mcp
npm install

setting

The MCP server settings are added to the configuration file of each AI assistant.

VSCode (for CLINE/Cursor)

~/Library/Application Support/Code/User/globalStorage/saoudrizwan.claude-dev/settings/cline_mcp_settings.json :

{
  "mcpServers": {
    "shared-knowledge-base": {
      "command": "node",
      "args": ["/path/to/shared-knowledge-mcp/dist/index.js"],
      "env": {
        "KNOWLEDGE_BASE_PATH": "/path/to/your/rules",
        "OPENAI_API_KEY": "your-openai-api-key",
        "SIMILARITY_THRESHOLD": "0.7",
        "CHUNK_SIZE": "1000",
        "CHUNK_OVERLAP": "200",
        "VECTOR_STORE_TYPE": "hnswlib"
      }
    }
  }
}

Examples of using Pinecone

{
  "mcpServers": {
    "shared-knowledge-base": {
      "command": "node",
      "args": ["/path/to/shared-knowledge-mcp/dist/index.js"],
      "env": {
        "KNOWLEDGE_BASE_PATH": "/path/to/your/rules",
        "OPENAI_API_KEY": "your-openai-api-key",
        "VECTOR_STORE_TYPE": "pinecone",
        "VECTOR_STORE_CONFIG": "{\"apiKey\":\"your-pinecone-api-key\",\"environment\":\"your-environment\",\"index\":\"your-index-name\"}"
      }
    }
  }
}

Claude Desktop

~/Library/Application Support/Claude/claude_desktop_config.json :

Example using HNSWLib (default)

{
  "mcpServers": {
    "shared-knowledge-base": {
      "command": "node",
      "args": ["/path/to/shared-knowledge-mcp/dist/index.js"],
      "env": {
        "KNOWLEDGE_BASE_PATH": "/path/to/your/docs",
        "OPENAI_API_KEY": "your-openai-api-key",
        "SIMILARITY_THRESHOLD": "0.7",
        "CHUNK_SIZE": "1000",
        "CHUNK_OVERLAP": "200",
        "VECTOR_STORE_TYPE": "hnswlib",
        "VECTOR_STORE_CONFIG": "{}"
      },
      "disabled": false,
      "autoApprove": []
    }
  }
}

Examples of using Weaviate

{
  "mcpServers": {
    "shared-knowledge-base": {
      "command": "node",
      "args": ["/path/to/shared-knowledge-mcp/dist/index.js"],
      "env": {
        "KNOWLEDGE_BASE_PATH": "/path/to/your/docs",
        "OPENAI_API_KEY": "your-openai-api-key",
        "SIMILARITY_THRESHOLD": "0.7",
        "CHUNK_SIZE": "1000",
        "CHUNK_OVERLAP": "200",
        "VECTOR_STORE_TYPE": "weaviate",
        "VECTOR_STORE_CONFIG": "{\"url\":\"http://localhost:8080\",\"className\":\"Document\",\"textKey\":\"content\"}"
      },
      "disabled": false,
      "autoApprove": []
    }
  }
}

Note : If you are using Weaviate, you must first start the Weaviate server, which can be done with the following command:

./start-weaviate.sh

development

Start the development server

npm run dev

Build

npm run build

Running in Production

npm start

Available Tools

rag_search

Search for information in the knowledge base.

Search Request

interface SearchRequest {
  // 検索クエリ（必須）
  query: string;
  
  // 返す結果の最大数（デフォルト: 5）
  limit?: number;
  
  // 検索のコンテキスト（オプション）
  context?: string;
  
  // フィルタリングオプション（オプション）
  filter?: {
    // ドキュメントの種類でフィルタリング（例: ["markdown", "code"]）
    documentTypes?: string[];
    
    // ソースパスのパターンでフィルタリング（例: "*.md"）
    sourcePattern?: string;
  };
  
  // 結果に含める情報（オプション）
  include?: {
    metadata?: boolean;   // メタデータを含める
    summary?: boolean;    // 要約を生成
    keywords?: boolean;   // キーワードを抽出
    relevance?: boolean;  // 関連性の説明を生成
  };
}

Usage Example

Basic search:

const result = await callTool("rag_search", {
  query: "コミットメッセージのフォーマット",
  limit: 3
});

Advanced Search:

const result = await callTool("rag_search", {
  query: "コミットメッセージのフォーマット",
  context: "Gitの使い方について調査中",
  filter: {
    documentTypes: ["markdown"],
    sourcePattern: "git-*.md"
  },
  include: {
    summary: true,
    keywords: true,
    relevance: true
  }
});

Search Results

interface SearchResult {
  // 検索クエリに関連する文書の内容
  content: string;
  
  // 類似度スコア（0-1）
  score: number;
  
  // ソースファイルのパス
  source: string;
  
  // 位置情報
  startLine?: number;     // 開始行
  endLine?: number;       // 終了行
  startColumn?: number;   // 開始桁
  endColumn?: number;     // 終了桁
  
  // ドキュメントの種類（例: "markdown", "code", "text"）
  documentType?: string;
  
  // 追加情報（include オプションで指定した場合のみ）
  summary?: string;       // コンテンツの要約
  keywords?: string[];    // 関連キーワード
  relevance?: string;     // 関連性の説明
  metadata?: Record<string, unknown>; // メタデータ
}

Response example

{
  "results": [
    {
      "content": "# コミットメッセージのフォーマット\n\n以下の形式でコミットメッセージを記述してください：\n\n```\n<type>(<scope>): <subject>\n\n<body>\n\n<footer>\n```\n\n...",
      "score": 0.92,
      "source": "/path/to/rules/git-conventions.md",
      "startLine": 1,
      "endLine": 10,
      "startColumn": 1,
      "endColumn": 35,
      "documentType": "markdown",
      "summary": "コミットメッセージのフォーマットについての説明文書",
      "keywords": ["commit", "message", "format", "type", "scope"],
      "relevance": "このドキュメントは検索クエリ \"コミットメッセージのフォーマット\" に関連する情報を含んでいます。類似度スコア: 0.92"
    }
  ]
}

These expanded search capabilities enable LLM to process information more accurately and efficiently. Additional information such as location, document type, abstract, and keywords help LLM to better understand and utilize search results.

structure

At startup, it reads Markdown files (.md, .mdx) and text files (.txt) in the specified directory.
Split the document into chunks and vectorize it using the OpenAI API
Creates a vector index using the selected vector store (default: HNSWLib)
Returns documents that are highly similar to a search query

Supported Vector Stores

HNSWLib : A fast vector store stored on the local file system (default)
Chroma : an open source vector database
Pinecone : Managed vector database service (API key required)
Milvus : A large-scale vector search engine
Weaviate : A schema-first vector database (Docker required)

Each vector store is exposed through an abstracted interface, making it easy to switch between them as needed.

How to navigate the Vector Store environment

HNSWLib (default)

HNSWLib saves the vector store on the local file system, so no special configuration is required.

Vector store reconstruction:

./rebuild-vector-store-hnsw.sh

Weaviate

To use Weaviate, you need Docker.

Start the Weaviate environment:

./start-weaviate.sh

Vector store reconstruction:

./rebuild-vector-store-weaviate.sh

Check the status of Weaviate:

curl http://localhost:8080/v1/.well-known/ready

Stopping the Weaviate environment:

docker-compose down

Delete your Weaviate data completely (only if necessary):

docker-compose down -v

Weaviate configuration is managed in the docker-compose.yml file. By default, the following settings are applied:

Port: 8080
Authentication: Anonymous access enabled
Vectorization module: None (use external padding)
Data storage: Docker volume ( weaviate_data )

Configuration options

environmental variables	explanation	Default value
KNOWLEDGE_BASE_PATH	Knowledge Base Path (required)	-
OPENAI_API_KEY	OpenAI API key (required)	-
SIMILARITY_THRESHOLD	Similarity score threshold for search (0-1)	0.7
CHUNK_SIZE	Chunk size for splitting text	1000
CHUNK_OVERLAP	Chunk overlap size	200
VECTOR_STORE_TYPE	The type of vector store to use ("hnswlib", "chroma", "pinecone", "milvus").	"hnswlib"
VECTOR_STORE_CONFIG	Vector store configuration (JSON string)	{}

license

ISC

contribution

Fork
Create a feature branch ( git checkout -b feature/amazing-feature )
Commit the changes ( git commit -m 'Add some amazing feature' )
Push to the branch ( git push origin feature/amazing-feature )
Create a Pull Request

Shared Knowledge MCP Server

Integrations