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

# Specification

> How to read RCP/1 — the section-by-section map into these docs, the normative document, and a complete worked session.

RCP/1 is **stable**. These documentation pages *are* the readable specification —
each protocol concept has its own page. This page is the map: it links every part
of the normative document to the page that explains it, states the conventions,
and shows a complete session end to end.

<CardGroup cols={2}>
  <Card title="Normative document" icon="scroll" href="https://github.com/1ay1/rcp/blob/main/spec/rcp-1.0.md">
    `spec/rcp-1.0.md` — the single source of truth, RFC-2119 + JSON-RPC 2.0.
  </Card>

  <Card title="JSON Schema" icon="brackets-curly" href="/reference/schema">
    Draft 2020-12 schema for every message shape.
  </Card>
</CardGroup>

## Conventions

The key words **MUST**, **MUST NOT**, **REQUIRED**, **SHOULD**, **SHOULD NOT**,
**MAY**, and **OPTIONAL** are used as defined in [RFC 2119](https://www.rfc-editor.org/rfc/rfc2119)
and [RFC 8174](https://www.rfc-editor.org/rfc/rfc8174) — and only when in **ALL
CAPS**. All messages are [JSON-RPC 2.0](https://www.jsonrpc.org/specification);
all bytes are UTF-8 JSON.

## Section map

| Spec §      | Topic                                    | Read it here                                                                         |
| ----------- | ---------------------------------------- | ------------------------------------------------------------------------------------ |
| §1–§2       | Goals, roles                             | [Introduction](/get-started/introduction), [Architecture](/get-started/architecture) |
| §3          | Retrieval pipeline model                 | [The retrieval pipeline](/concepts/pipeline)                                         |
| §4          | Message format, `_meta`, Content blocks  | [Message Format](/concepts/message-format)                                           |
| §5          | Transports (stdio, HTTP, SSE)            | [Transports](/concepts/transports)                                                   |
| §6          | Capabilities                             | [Capabilities](/concepts/capabilities)                                               |
| §7.1        | `initialize` / `info`                    | [Initialization](/concepts/initialization)                                           |
| §7.3        | `embed` / `embed/sparse` / `embed/multi` | [Embedding](/methods/embed)                                                          |
| §7.5        | `rerank`                                 | [Rerank](/methods/rerank)                                                            |
| §7.7        | `retrieve`                               | [Retrieve](/methods/retrieve)                                                        |
| §7.8        | `query/transform`                        | [Query Transform](/methods/query-transform)                                          |
| §7.9        | `graph`                                  | [Graph](/methods/graph)                                                              |
| §7.10–§7.11 | `index/add` / `index/delete`             | [Indexing](/methods/index)                                                           |
| §7.12       | `catalog/list`                           | [Catalog](/methods/catalog)                                                          |
| §7.14–§7.15 | `cancel`, `ping`, `shutdown`             | [Lifecycle](/concepts/lifecycle)                                                     |
| §8          | Metadata filtering                       | [Metadata Filtering](/features/metadata-filtering)                                   |
| §9          | Streaming & progress                     | [Streaming & Progress](/features/streaming-progress)                                 |
| §10         | Pagination                               | [Pagination](/features/pagination)                                                   |
| §11         | Batching                                 | [Batching](/features/batching)                                                       |
| §12         | Errors                                   | [Errors](/concepts/errors)                                                           |
| §13         | Session lifecycle                        | [Lifecycle](/concepts/lifecycle)                                                     |
| §14         | Conformance                              | [Conformance](/operating/conformance)                                                |
| §15         | Security                                 | [Security](/operating/security)                                                      |
| §16         | Federation                               | [Federation & Selector](/features/federation)                                        |
| §17         | Observability                            | [Observability](/features/observability)                                             |

## Mapping SOTA techniques to RCP

Every mainstream retrieval technique maps onto an RCP surface — the protocol was
designed backwards from the state of the art:

| Technique                                      | RCP surface                                                       |
| ---------------------------------------------- | ----------------------------------------------------------------- |
| BM25 / lexical                                 | `retrieve mode:"sparse"`                                          |
| Dense bi-encoder (e5/BGE/GTE)                  | `retrieve mode:"dense"`, or `embed` for client-side ANN           |
| Learned-sparse (SPLADE)                        | `sparseEmbed` + `retrieve mode:"sparse"`                          |
| Hybrid + RRF                                   | `retrieve mode:"hybrid", fusion:{method:"rrf"}`                   |
| Cross-encoder rerank                           | `rerank method:"cross-encoder"` or `retrieve.rerank`              |
| Late interaction (ColBERT)                     | `multiVector` + `rerank method:"colbert"`                         |
| Visual-document retrieval (ColPali/ColQwen)    | `multiVector modalities:["image"]`, `image` Content blocks        |
| Matryoshka / truncatable embeddings            | `embed.dimension` advertised; client truncates                    |
| MMR diversification                            | `retrieve.mmr`                                                    |
| Query rewrite / HyDE / multi-query / step-back | `query/transform` or `retrieve.rewrite`                           |
| Multi-hop / agentic                            | client loop over `query/transform` + `retrieve` + `graph:"drift"` |
| GraphRAG local/global                          | `graph op:"local"\|"global"`                                      |
| Contextual retrieval (Anthropic)               | `index/add contextual:true`                                       |
| Citations / provenance                         | `citations`, `Hit.citation`, `Hit.trust`                          |
| Freshness / recency                            | `retrieve.recency`                                                |
| Reproducible eval                              | `retrieve.seed`, `retrieve.indexVersion`                          |
| Multi-backend fan-out + fusion                 | [Federation](/features/federation), RRF / weighted                |

## A worked session

A complete stdio session against a SOTA server — handshake, a hybrid + rerank +
MMR retrieval with a filter and a progress stream, a cancelled call, and
shutdown. `→` is client→server, `←` is server→client; one JSON object per line.

```json theme={null}
→ {"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":1,"client":{"name":"my-app","version":"0.4"}}}
← {"jsonrpc":"2.0","id":1,"result":{"protocolVersion":1,"server":{"name":"acme-rag","version":"2.1"},"capabilities":{"retrieve":{"maxK":200,"modes":["dense","sparse","hybrid"],"citations":true},"rerank":{"methods":["cross-encoder","colbert"]},"filter":{"fields":{"year":"int","lang":"keyword"}},"streaming":{}}}}

→ {"jsonrpc":"2.0","id":2,"method":"retrieve","params":{"query":"transformer attention complexity","k":5,"mode":"hybrid","rerank":{"method":"cross-encoder","topN":100},"mmr":{"lambda":0.5},"filter":{"and":[{"field":"lang","op":"eq","value":"en"},{"field":"year","op":"gte","value":2017}]},"_meta":{"progressToken":"t2"}}}
← {"jsonrpc":"2.0","method":"notifications/progress","params":{"progressToken":"t2","progress":0.4,"message":"recall"}}
← {"jsonrpc":"2.0","method":"notifications/progress","params":{"progressToken":"t2","progress":0.8,"message":"rerank"}}
← {"jsonrpc":"2.0","id":2,"result":{"hits":[{"id":"arxiv:1706.03762#3","score":0.94,"text":"Attention is all you need …","citation":{"uri":"https://arxiv.org/abs/1706.03762","title":"Attention Is All You Need"},"meta":{"year":2017}}]}}

→ {"jsonrpc":"2.0","id":3,"method":"graph","params":{"op":"global","query":"survey of efficient attention"}}
→ {"jsonrpc":"2.0","method":"notifications/cancel","params":{"id":3,"reason":"user navigated away"}}
← {"jsonrpc":"2.0","id":3,"error":{"code":-32006,"message":"cancelled"}}

→ {"jsonrpc":"2.0","id":4,"method":"shutdown","params":{}}
← {"jsonrpc":"2.0","id":4,"result":{}}
```

<Info>
  For evaluation and quality (Recall\@k, MRR, nDCG\@k, MAP) and the reproducibility
  handles behind them, see the normative
  [Appendix F](https://github.com/1ay1/rcp/blob/main/spec/rcp-1.0.md#appendix-f--evaluation--quality).
</Info>