> ## Documentation Index
> Fetch the complete documentation index at: https://retrievalcontextprotocol.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Query Transform

> Rewrite, expand, or decompose a query — HyDE, multi-query, decomposition, and step-back — for clients driving an agentic loop.

`query/transform` is gated by **`transform`**. It exposes query transformation as
a standalone step so a client orchestrating its own agentic retrieval loop can
plan queries before recall. (A server can also apply the same transforms inside
[`retrieve`](/methods/retrieve) via the `rewrite` option.)

<CodeGroup>
  ```json Request theme={null}
  {
    "jsonrpc": "2.0", "id": 1, "method": "query/transform",
    "params": { "query": "how do transformers scale?", "method": "multi-query", "n": 3 }
  }
  ```

  ```json Response theme={null}
  {
    "jsonrpc": "2.0", "id": 1, "result": {
      "queries": [
        "computational complexity of transformer attention",
        "efficient transformer architectures for long sequences",
        "scaling laws for transformer models"
      ]
    }
  }
  ```
</CodeGroup>

## Params

<ParamField path="query" type="string" required>
  The original query.
</ParamField>

<ParamField path="method" type="&#x22;rewrite&#x22; | &#x22;hyde&#x22; | &#x22;multi-query&#x22; | &#x22;decompose&#x22; | &#x22;step-back&#x22;" required>
  The transformation to apply. **MUST** be one of the advertised
  `transform.methods`.
</ParamField>

<ParamField path="n" type="integer">
  How many queries to produce (for `multi-query` / `decompose`).
</ParamField>

## Result

<ResponseField name="queries" type="string[]">
  The transformed queries. `multi-query`/`decompose` return several; `rewrite`
  and `step-back` return one.
</ResponseField>

<ResponseField name="hypothetical" type="string">
  For `hyde`, a hypothetical answer document the client then **embeds** and uses
  as the search vector.
</ResponseField>

## The transformations

<CardGroup cols={2}>
  <Card title="rewrite" icon="pen">
    Clean up and disambiguate a raw user query.
  </Card>

  <Card title="hyde" icon="lightbulb">
    Generate a hypothetical answer; embed *it* instead of the question
    (Hypothetical Document Embeddings).
  </Card>

  <Card title="multi-query" icon="clone">
    Produce several paraphrases to widen recall, then fuse the result sets.
  </Card>

  <Card title="decompose" icon="diagram-next">
    Break a complex question into answerable sub-questions.
  </Card>

  <Card title="step-back" icon="arrow-left">
    Ask a more general "step-back" question to retrieve grounding principles.
  </Card>
</CardGroup>