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

# Graph

> GraphRAG — local entity-centric retrieval, global community summaries, and drift search over a knowledge graph.

`graph` is gated by **`graph`**. It exposes GraphRAG-style retrieval over a
knowledge graph the server maintains, selected by an `op`. Servers advertise the
ops they support in `graph.ops`.

<CodeGroup>
  ```json Request theme={null}
  {
    "jsonrpc": "2.0", "id": 1, "method": "graph",
    "params": { "op": "local", "query": "who influenced Alan Turing?", "k": 5 }
  }
  ```

  ```json Response theme={null}
  { "jsonrpc": "2.0", "id": 1, "result": { "hits": [ { "id": "e:turing", "score": 0.9, "text": "…" } ] } }
  ```
</CodeGroup>

## Params

<ParamField path="op" type="&#x22;local&#x22; | &#x22;global&#x22; | &#x22;drift&#x22; | &#x22;communities&#x22; | &#x22;neighbors&#x22;" required>
  The graph operation. **MUST** be one of the advertised `graph.ops`.
</ParamField>

<ParamField path="query" type="string">
  The natural-language query (for `local`, `global`, `drift`).
</ParamField>

<ParamField path="k" type="integer">
  Result count where applicable.
</ParamField>

<ParamField path="node" type="string | number">
  A node id (for `neighbors`).
</ParamField>

## Operations

<CardGroup cols={2}>
  <Card title="local" icon="location-dot">
    Entity-centric retrieval: find the entities and relationships closest to the
    query and gather their neighbourhood. Returns `hits`.
  </Card>

  <Card title="global" icon="earth-americas">
    Map-reduce over precomputed **community summaries** for corpus-wide,
    thematic "sensemaking" questions. Returns a `summary` and `communities`.
  </Card>

  <Card title="drift" icon="route">
    Dynamic Reasoning and Inference over a Fuzzy Trace — blends local and global,
    following the graph as reasoning unfolds.
  </Card>

  <Card title="communities / neighbors" icon="circle-nodes">
    `communities` lists detected communities; `neighbors` expands the adjacency
    of a given `node`.
  </Card>
</CardGroup>

## Result

Shapes depend on the op. Entity-centric ops (`local`, `drift`, `neighbors`)
return **`hits`** with the same [Hit shape](/methods/retrieve#the-hit-shape) as
`retrieve`. Corpus-wide ops (`global`, `communities`) return a **`summary`**
string and a **`communities`** array.

<CodeGroup>
  ```json op: "global" theme={null}
  { "result": {
      "summary": "The corpus centres on three research communities …",
      "communities": [ { "id": "c0", "title": "Deep learning", "size": 214 } ] } }
  ```
</CodeGroup>

<Info>
  GraphRAG unifies local (entity) and global (thematic) retrieval under one method
  so a client can ask both "what connects X and Y?" and "what are the big themes?"
  without a second protocol.
</Info>