What is Semantic search?

Semantic search retrieves information based on what text means instead of which words it literally contains. The technique works by turning both the query and the candidate documents into embeddings, vectors that place semantically similar content near each other in a high-dimensional space, and then returning the documents whose vectors are closest to the query's. Because it matches on meaning, semantic search handles synonyms, paraphrases, and conceptual questions that keyword search misses: a search for how do I cancel my plan can surface a page titled subscription termination even with no shared words. It is the retrieval half of RAG and is typically backed by a vector database that performs the nearest-neighbor lookup. Many real systems combine it with keyword (lexical) search in a hybrid approach, since exact terms like error codes or product names still matter. In an agent setting, semantic search is frequently exposed as an MCP tool, over a documentation set, a knowledge base, or a memory store, so the model can pull in conceptually relevant context on demand rather than relying only on what it was trained on.