mcp-documentation-server

Semantic search over local documents works by indexing your uploaded files into an embedded Orama vector database, applying parent-child chunking to preserve context so search results return surrounding text sections for accurate retrieval.

To build a searchable knowledge base from multiple files, configure the MCP documentation server in your client, upload your documents, and the system automatically indexes them on disk for fast repeat searches without external dependencies.

You do not need a Gemini API key for vector database search. A GEMINI_API_KEY is optional and only required if you want to enable AI-powered, higher-level question answering with citations over your documents.

The best way to search internal runbooks with exact and conceptual queries is using hybrid full-text matching combined with semantic vector search, which improves relevance for both specific terms and broader contextual questions.

Yes, you can use parent-child chunking to provide context windows for an LLM. The chunking mechanism returns surrounding document sections alongside search hits, giving the LLM necessary context to answer accurately.