vector_search_hash

Find similar vectors stored in Redis hash structures using K-nearest neighbor search to retrieve relevant documents based on vector similarity.

Instructions

Perform a KNN vector similarity search using Redis 8 or later version on vectors stored in hash data structures.

Args: query_vector: List of floats to use as the query vector. index_name: Name of the Redis index. Unless specifically specified, use the default index name. vector_field: Name of the indexed vector field. Unless specifically required, use the default field name k: Number of nearest neighbors to return. return_fields: List of fields to return (optional).

Returns: A list of matched documents or an error message.

Input Schema

TableJSON Schema

Name	Required	Default
`query_vector`	Yes
`index_name`	No	vector_index
`vector_field`	No	vector
`k`	No
`return_fields`	No

Implementation Reference

src/tools/redis_query_engine.py:104-150 (handler)

The core handler function for the 'vector_search_hash' MCP tool. It performs KNN vector similarity search on Redis hashes using HNSW index. Decorated with @mcp.tool() for automatic registration.

@mcp.tool()
async def vector_search_hash(
    query_vector: List[float],
    index_name: str = "vector_index",
    vector_field: str = "vector",
    k: int = 5,
    return_fields: Optional[List[str]] = None,
) -> Union[List[Dict[str, Any]], str]:
    """
    Perform a KNN vector similarity search using Redis 8 or later version on vectors stored in hash data structures.

    Args:
        query_vector: List of floats to use as the query vector.
        index_name: Name of the Redis index. Unless specifically specified, use the default index name.
        vector_field: Name of the indexed vector field. Unless specifically required, use the default field name
        k: Number of nearest neighbors to return.
        return_fields: List of fields to return (optional).

    Returns:
        A list of matched documents or an error message.
    """
    try:
        r = RedisConnectionManager.get_connection()

        # Convert query vector to float32 binary blob
        vector_blob = np.array(query_vector, dtype=np.float32).tobytes()

        # Build the KNN query
        base_query = f"*=>[KNN {k} @{vector_field} $vec_param AS score]"
        query = (
            Query(base_query)
            .sort_by("score")
            .paging(0, k)
            .return_fields("id", "score", *return_fields or [])
            .dialect(2)
        )

        # Perform the search with vector parameter
        results = r.ft(index_name).search(
            query, query_params={"vec_param": vector_blob}
        )

        # Format and return the results
        return [doc.__dict__ for doc in results.docs]
    except RedisError as e:
        return f"Error performing vector search on index '{index_name}': {str(e)}"

src/tools/redis_query_engine.py:104-104 (registration)
The @mcp.tool() decorator registers the vector_search_hash function as an MCP tool.
```
@mcp.tool()
```

src/tools/redis_query_engine.py:105-111 (schema)

Function signature provides the input schema (parameters with types and defaults) and output type for the tool.

async def vector_search_hash(
    query_vector: List[float],
    index_name: str = "vector_index",
    vector_field: str = "vector",
    k: int = 5,
    return_fields: Optional[List[str]] = None,
) -> Union[List[Dict[str, Any]], str]:

Redis MCP Server