MCP RAG Tool

Provides semantic document search through the RAG pipeline. Searches uploaded documents for content relevant to the query.

• Tech: TypeScript, MCP SDK
• Port: 3002
• Tool name: search_documents

Tool: search_documents

Accepts 1-3 query candidates ordered by specificity. Tries each query sequentially until results are found.

Input Schema

Field	Type	Required	Description
queries	string[]	Yes	1-3 search queries, ordered from most to least specific
similarityTopK	number	No	Number of similarity search results
scoreThreshold	number	No	Minimum similarity score
rerankTopK	number	No	Number of results after reranking
rerankScore	number	No	Minimum rerank score

Example

{
  "name": "search_documents",
  "arguments": {
    "queries": [
      "kubernetes pod scheduling algorithm",
      "kubernetes scheduling",
      "container orchestration"
    ],
    "similarityTopK": 15,
    "rerankTopK": 5
  }
}

How It Works

1. Takes the first query from the list
2. Calls the RAG service to perform similarity search + reranking
3. If results meet the score threshold, returns them
4. If not, tries the next query
5. Returns the best results found

Connection

Connects to the RAG service backend to perform the actual retrieval. The RAG service URL is configured via environment variables.