Anthropic Claude Messages Streaming API

asyncapi: 2.6.0
id: urn:com:anthropic:claude:messages:streaming
info:
  title: Anthropic Claude Messages Streaming API
  version: '2023-06-01'
  description: |
    AsyncAPI specification modeling the Server-Sent Events (SSE) stream produced by
    Anthropic's Claude Messages API when `"stream": true` is set on a POST to
    `/v1/messages`.

    Transport: HTTP/1.1 with `Content-Type: text/event-stream` (Server-Sent Events).
    The client opens a long-lived HTTP POST and the server pushes a sequence of
    named SSE events. Each event has an SSE event name (e.g. `event: message_stop`)
    and a `data:` line containing a JSON object whose `type` field matches the
    event name.

    Event flow per stream:
      1. `message_start` — opens the stream with a Message object (empty content).
      2. For each content block: `content_block_start`, one or more
         `content_block_delta` events, then `content_block_stop`.
      3. One or more `message_delta` events with top-level Message updates
         (cumulative token usage in `usage`).
      4. A final `message_stop` event.

    `ping` events may appear at any point. `error` events may appear during
    periods of high load or other failure conditions.

    `content_block_delta` carries a `delta` whose `type` is one of:
      - `text_delta` — incremental text for text content blocks
      - `input_json_delta` — partial JSON string for `tool_use` / `server_tool_use`
      - `thinking_delta` — incremental thinking content for extended-thinking blocks
      - `signature_delta` — final signature for a thinking block (sent just before
        `content_block_stop`)

    This specification is derived verbatim from the public Anthropic documentation
    at https://docs.anthropic.com/en/api/messages-streaming (currently served from
    https://platform.claude.com/docs/en/api/messages-streaming).
  contact:
    name: Anthropic Developer Support
    url: https://docs.anthropic.com/en/api/getting-help
  license:
    name: Anthropic Commercial Terms of Service
    url: https://www.anthropic.com/legal/commercial-terms
  termsOfService: https://www.anthropic.com/legal/aup
defaultContentType: text/event-stream
servers:
  production:
    url: api.anthropic.com/v1
    protocol: https
    description: |
      Anthropic public REST + SSE base URL. Streaming responses are delivered
      over HTTP/1.1 with `Content-Type: text/event-stream` (HTTP+SSE transport).
      Standard request headers apply: `x-api-key`, `anthropic-version`,
      `content-type: application/json`.
    bindings:
      http:
        bindingVersion: '0.3.0'
channels:
  /messages:
    description: |
      Server-Sent Events stream returned by `POST https://api.anthropic.com/v1/messages`
      when the request body includes `"stream": true`. Each SSE frame's event name
      maps to one of the messages defined below; the JSON payload after `data:`
      carries the matching schema.
    bindings:
      http:
        type: request
        method: POST
        bindingVersion: '0.3.0'
    subscribe:
      operationId: subscribeMessagesStream
      summary: Receive Anthropic Claude Messages streaming events over HTTP+SSE
      description: |
        After issuing `POST /v1/messages` with `stream: true`, the client reads
        an unbounded sequence of SSE events from the same HTTP response body
        until `message_stop` (or `error`) is received and the server closes
        the connection.
      bindings:
        http:
          bindingVersion: '0.3.0'
      message:
        oneOf:
          - $ref: '#/components/messages/MessageStart'
          - $ref: '#/components/messages/ContentBlockStart'
          - $ref: '#/components/messages/ContentBlockDelta'
          - $ref: '#/components/messages/ContentBlockStop'
          - $ref: '#/components/messages/MessageDelta'
          - $ref: '#/components/messages/MessageStop'
          - $ref: '#/components/messages/Ping'
          - $ref: '#/components/messages/Error'
components:
  messages:
    MessageStart:
      name: message_start
      title: message_start
      summary: Opens a streamed response with a Message object whose content array is empty.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/MessageStartEvent'
      examples:
        - name: basicMessageStart
          summary: message_start event from the Basic streaming request example
          payload:
            type: message_start
            message:
              id: msg_1nZdL29xx5MUA1yADyHTEsnR8uuvGzszyY
              type: message
              role: assistant
              content: []
              model: claude-opus-4-8
              stop_reason: null
              stop_sequence: null
              usage:
                input_tokens: 25
                output_tokens: 1
    ContentBlockStart:
      name: content_block_start
      title: content_block_start
      summary: Marks the start of a content block at a given index in the final Message.content array.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/ContentBlockStartEvent'
      examples:
        - name: textBlockStart
          payload:
            type: content_block_start
            index: 0
            content_block:
              type: text
              text: ''
        - name: toolUseBlockStart
          payload:
            type: content_block_start
            index: 1
            content_block:
              type: tool_use
              id: toolu_01T1x1fJ34qAmk2tNTrN7Up6
              name: get_weather
              input: {}
        - name: thinkingBlockStart
          payload:
            type: content_block_start
            index: 0
            content_block:
              type: thinking
              thinking: ''
              signature: ''
        - name: serverToolUseBlockStart
          payload:
            type: content_block_start
            index: 1
            content_block:
              type: server_tool_use
              id: srvtoolu_014hJH82Qum7Td6UV8gDXThB
              name: web_search
              input: {}
    ContentBlockDelta:
      name: content_block_delta
      title: content_block_delta
      summary: Carries a partial update for the content block at the specified index.
      description: |
        The shape of `delta` is determined by `delta.type`:
          - `text_delta`        — `delta.text` (string fragment)
          - `input_json_delta`  — `delta.partial_json` (partial JSON string)
          - `thinking_delta`    — `delta.thinking` (string fragment)
          - `signature_delta`   — `delta.signature` (final signature string)
      contentType: application/json
      payload:
        $ref: '#/components/schemas/ContentBlockDeltaEvent'
      examples:
        - name: textDelta
          summary: Text delta example from the docs
          payload:
            type: content_block_delta
            index: 0
            delta:
              type: text_delta
              text: ello frien
        - name: inputJsonDelta
          summary: Input JSON delta example from the docs
          payload:
            type: content_block_delta
            index: 1
            delta:
              type: input_json_delta
              partial_json: '{"location": "San Fra'
        - name: thinkingDelta
          summary: Thinking delta example from the docs
          payload:
            type: content_block_delta
            index: 0
            delta:
              type: thinking_delta
              thinking: |-
                I need to find the GCD of 1071 and 462 using the Euclidean algorithm.

                1071 = 2 × 462 + 147
        - name: signatureDelta
          summary: Signature delta example from the docs
          payload:
            type: content_block_delta
            index: 0
            delta:
              type: signature_delta
              signature: EqQBCgIYAhIM1gbcDa9GJwZA2b3hGgxBdjrkzLoky3dl1pkiMOYds...
    ContentBlockStop:
      name: content_block_stop
      title: content_block_stop
      summary: Marks the end of the content block at the given index.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/ContentBlockStopEvent'
      examples:
        - name: contentBlockStop
          payload:
            type: content_block_stop
            index: 0
    MessageDelta:
      name: message_delta
      title: message_delta
      summary: Top-level updates to the final Message (stop reason, sequence) plus cumulative usage.
      description: |
        The token counts shown in the `usage` field of `message_delta` events are
        cumulative across the stream.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/MessageDeltaEvent'
      examples:
        - name: basicMessageDelta
          payload:
            type: message_delta
            delta:
              stop_reason: end_turn
              stop_sequence: null
            usage:
              output_tokens: 15
        - name: toolUseMessageDelta
          payload:
            type: message_delta
            delta:
              stop_reason: tool_use
              stop_sequence: null
            usage:
              output_tokens: 89
        - name: webSearchMessageDelta
          payload:
            type: message_delta
            delta:
              stop_reason: end_turn
              stop_sequence: null
            usage:
              input_tokens: 10682
              cache_creation_input_tokens: 0
              cache_read_input_tokens: 0
              output_tokens: 510
              server_tool_use:
                web_search_requests: 1
    MessageStop:
      name: message_stop
      title: message_stop
      summary: Final event on the stream; the connection is closed afterward.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/MessageStopEvent'
      examples:
        - name: messageStop
          payload:
            type: message_stop
    Ping:
      name: ping
      title: ping
      summary: Keep-alive event; may appear any number of times during the stream.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/PingEvent'
      examples:
        - name: ping
          payload:
            type: ping
    Error:
      name: error
      title: error
      summary: Error event emitted in the stream (e.g. overloaded_error during high load).
      description: |
        Errors that would normally surface as an HTTP error in non-streaming mode
        are emitted as `error` SSE events while the stream is open. The `error.type`
        field uses the same identifiers as the standard Anthropic Errors API
        (see https://docs.anthropic.com/en/api/errors).
      contentType: application/json
      payload:
        $ref: '#/components/schemas/ErrorEvent'
      examples:
        - name: overloadedError
          summary: Example error from the docs
          payload:
            type: error
            error:
              type: overloaded_error
              message: Overloaded
  schemas:
    MessageStartEvent:
      type: object
      required:
        - type
        - message
      properties:
        type:
          type: string
          const: message_start
          enum: [message_start]
        message:
          $ref: '#/components/schemas/Message'
    ContentBlockStartEvent:
      type: object
      required:
        - type
        - index
        - content_block
      properties:
        type:
          type: string
          const: content_block_start
          enum: [content_block_start]
        index:
          type: integer
          minimum: 0
          description: Index of this content block in the final Message.content array.
        content_block:
          $ref: '#/components/schemas/ContentBlock'
    ContentBlockDeltaEvent:
      type: object
      required:
        - type
        - index
        - delta
      properties:
        type:
          type: string
          const: content_block_delta
          enum: [content_block_delta]
        index:
          type: integer
          minimum: 0
        delta:
          oneOf:
            - $ref: '#/components/schemas/TextDelta'
            - $ref: '#/components/schemas/InputJsonDelta'
            - $ref: '#/components/schemas/ThinkingDelta'
            - $ref: '#/components/schemas/SignatureDelta'
          discriminator:
            propertyName: type
    ContentBlockStopEvent:
      type: object
      required:
        - type
        - index
      properties:
        type:
          type: string
          const: content_block_stop
          enum: [content_block_stop]
        index:
          type: integer
          minimum: 0
    MessageDeltaEvent:
      type: object
      required:
        - type
        - delta
      properties:
        type:
          type: string
          const: message_delta
          enum: [message_delta]
        delta:
          type: object
          description: Top-level Message fields that changed.
          properties:
            stop_reason:
              type: [string, 'null']
              enum: [end_turn, max_tokens, stop_sequence, tool_use, pause_turn, refusal, null]
              description: Reason the model stopped generating. May be null mid-stream.
            stop_sequence:
              type: [string, 'null']
        usage:
          $ref: '#/components/schemas/StreamingUsage'
    MessageStopEvent:
      type: object
      required:
        - type
      properties:
        type:
          type: string
          const: message_stop
          enum: [message_stop]
    PingEvent:
      type: object
      required:
        - type
      properties:
        type:
          type: string
          const: ping
          enum: [ping]
    ErrorEvent:
      type: object
      required:
        - type
        - error
      properties:
        type:
          type: string
          const: error
          enum: [error]
        error:
          type: object
          required:
            - type
            - message
          properties:
            type:
              type: string
              description: Anthropic error type identifier (e.g. overloaded_error).
              examples:
                - overloaded_error
                - invalid_request_error
                - authentication_error
                - permission_error
                - not_found_error
                - request_too_large
                - rate_limit_error
                - api_error
            message:
              type: string
              description: Human-readable error message.

    TextDelta:
      type: object
      description: Incremental text appended to a `text` content block.
      required:
        - type
        - text
      properties:
        type:
          type: string
          const: text_delta
          enum: [text_delta]
        text:
          type: string
    InputJsonDelta:
      type: object
      description: |
        Partial JSON string update for the `input` field of a `tool_use` or
        `server_tool_use` content block. Concatenate `partial_json` chunks
        across deltas to reconstruct the final JSON object on
        `content_block_stop`.
      required:
        - type
        - partial_json
      properties:
        type:
          type: string
          const: input_json_delta
          enum: [input_json_delta]
        partial_json:
          type: string
          description: A partial JSON string fragment.
    ThinkingDelta:
      type: object
      description: Incremental thinking content for an extended-thinking block.
      required:
        - type
        - thinking
      properties:
        type:
          type: string
          const: thinking_delta
          enum: [thinking_delta]
        thinking:
          type: string
    SignatureDelta:
      type: object
      description: |
        Final signature for a `thinking` content block. Emitted exactly once,
        immediately before the block's `content_block_stop`. Used to verify
        the integrity of the thinking block.
      required:
        - type
        - signature
      properties:
        type:
          type: string
          const: signature_delta
          enum: [signature_delta]
        signature:
          type: string

    Message:
      type: object
      description: Anthropic Message object (as emitted at stream open with empty content).
      required:
        - id
        - type
        - role
        - content
        - model
      properties:
        id:
          type: string
        type:
          type: string
          const: message
          enum: [message]
        role:
          type: string
          const: assistant
          enum: [assistant]
        content:
          type: array
          items:
            $ref: '#/components/schemas/ContentBlock'
        model:
          type: string
        stop_reason:
          type: [string, 'null']
          enum: [end_turn, max_tokens, stop_sequence, tool_use, pause_turn, refusal, null]
        stop_sequence:
          type: [string, 'null']
        usage:
          $ref: '#/components/schemas/StreamingUsage'
    ContentBlock:
      oneOf:
        - $ref: '#/components/schemas/TextBlock'
        - $ref: '#/components/schemas/ToolUseBlock'
        - $ref: '#/components/schemas/ServerToolUseBlock'
        - $ref: '#/components/schemas/ThinkingBlock'
        - $ref: '#/components/schemas/WebSearchToolResultBlock'
      discriminator:
        propertyName: type
    TextBlock:
      type: object
      required:
        - type
        - text
      properties:
        type:
          type: string
          const: text
          enum: [text]
        text:
          type: string
    ToolUseBlock:
      type: object
      required:
        - type
        - id
        - name
        - input
      properties:
        type:
          type: string
          const: tool_use
          enum: [tool_use]
        id:
          type: string
        name:
          type: string
        input:
          type: object
          description: Tool input object. Starts empty; the final value is the JSON object
            reconstructed by concatenating `input_json_delta.partial_json` chunks.
    ServerToolUseBlock:
      type: object
      required:
        - type
        - id
        - name
        - input
      properties:
        type:
          type: string
          const: server_tool_use
          enum: [server_tool_use]
        id:
          type: string
        name:
          type: string
        input:
          type: object
    ThinkingBlock:
      type: object
      required:
        - type
        - thinking
        - signature
      properties:
        type:
          type: string
          const: thinking
          enum: [thinking]
        thinking:
          type: string
        signature:
          type: string
    WebSearchToolResultBlock:
      type: object
      required:
        - type
        - tool_use_id
        - content
      properties:
        type:
          type: string
          const: web_search_tool_result
          enum: [web_search_tool_result]
        tool_use_id:
          type: string
        content:
          type: array
          items:
            type: object
            properties:
              type:
                type: string
              title:
                type: string
              url:
                type: string
              encrypted_content:
                type: string
              page_age:
                type: [string, 'null']
    StreamingUsage:
      type: object
      description: |
        Token usage object. Counts reported on `message_delta` events are
        cumulative across the stream.
      properties:
        input_tokens:
          type: integer
          minimum: 0
        output_tokens:
          type: integer
          minimum: 0
        cache_creation_input_tokens:
          type: integer
          minimum: 0
        cache_read_input_tokens:
          type: integer
          minimum: 0
        server_tool_use:
          type: object
          properties:
            web_search_requests:
              type: integer
              minimum: 0