> ## Documentation Index > Fetch the complete documentation index at: https://developers.meshapi.ai/llms.txt > Use this file to discover all available pages before exploring further. # Create Message > Anthropic-compatible Messages endpoint. Auth: x-api-key: rsk_ (or Authorization: Bearer rsk_) Models: any model this gateway serves, not only Anthropic's Streaming: set stream=true for Anthropic SSE events ## OpenAPI ````yaml /api/openapi.json post /v1/messages openapi: 3.1.0 info: title: MeshAPI description: One key, all AI models. version: 0.1.0 servers: - url: https://api.meshapi.ai security: - BearerAuth: [] paths: /v1/messages: post: tags: - Messages summary: Create Message description: >- Anthropic-compatible Messages endpoint. Auth: x-api-key: rsk_ (or Authorization: Bearer rsk_) Models: any model this gateway serves, not only Anthropic's Streaming: set stream=true for Anthropic SSE events operationId: create_message parameters: - name: x-api-key in: header required: false schema: anyOf: - type: string - type: 'null' title: X-Api-Key - in: header name: X-Mesh-Version required: false schema: type: string enum: - 2026-08 default: 2026-08 example: 2026-08 description: >- Dated version of the API contract to pin this request to. Omit it and the request is served under `2026-08` — the oldest supported version, so an existing integration is never moved by a release. A malformed or unsupported value is rejected with `400 invalid_api_version` rather than falling back silently. The version actually served is echoed as `X-Mesh-Version` on every response, including errors. requestBody: required: true content: application/json: schema: $ref: '#/components/schemas/AnthropicMessagesRequest' responses: '200': description: Successful Response content: application/json: schema: {} '422': description: Validation Error content: application/json: schema: $ref: '#/components/schemas/HTTPValidationError' security: - BearerAuth: [] components: schemas: AnthropicMessagesRequest: properties: model: type: string title: Model max_tokens: type: integer minimum: 1 title: Max Tokens messages: items: $ref: '#/components/schemas/AnthropicMessage' type: array title: Messages system: anyOf: - type: string - items: $ref: '#/components/schemas/TextBlock' type: array - type: 'null' title: System temperature: anyOf: - type: number maximum: 1 minimum: 0 - type: 'null' title: Temperature top_p: anyOf: - type: number maximum: 1 minimum: 0 - type: 'null' title: Top P stop_sequences: anyOf: - items: type: string type: array - type: 'null' title: Stop Sequences stream: type: boolean title: Stream default: false metadata: anyOf: - additionalProperties: true type: object - type: 'null' title: Metadata tools: anyOf: - items: $ref: '#/components/schemas/AnthropicTool' type: array - type: 'null' title: Tools tool_choice: anyOf: - additionalProperties: true type: object - type: 'null' title: Tool Choice top_k: anyOf: - type: integer - type: 'null' title: Top K thinking: anyOf: - additionalProperties: true type: object - type: 'null' title: Thinking type: object required: - model - max_tokens - messages title: AnthropicMessagesRequest example: max_tokens: 1024 messages: - content: say hi role: user model: anthropic/claude-haiku-4.5 HTTPValidationError: properties: detail: items: $ref: '#/components/schemas/ValidationError' type: array title: Detail type: object title: HTTPValidationError AnthropicMessage: properties: role: type: string enum: - user - assistant - system title: Role content: anyOf: - type: string - items: anyOf: - $ref: '#/components/schemas/TextBlock' - $ref: '#/components/schemas/ImageBlock' - $ref: '#/components/schemas/ToolUseBlock' - $ref: '#/components/schemas/ToolResultBlock' type: array title: Content type: object required: - role - content title: AnthropicMessage description: >- One turn. Anthropic's documented roles are ``user`` and ``assistant`` — system text is a top-level field, which is the biggest structural difference from the OpenAI shape. But ``system`` is accepted here as a THIRD role, because Claude Code actually sends one: a request captured against a real session carried ``roles: ["user", "system"]`` alongside a populated top-level ``system``. Rejecting it is not "being strict about the spec" — it is being incompatible with the single client this endpoint exists to serve, and it fails on the very first request with a 400 that names an internal field path. OpenAI has a native ``system`` role, so the mapping is direct and lossless. TextBlock: properties: type: type: string const: text title: Type default: text text: type: string title: Text cache_control: anyOf: - additionalProperties: true type: object - type: 'null' title: Cache Control type: object required: - text title: TextBlock description: 'A ``{"type": "text", "text": ...}`` content block.' AnthropicTool: properties: name: type: string title: Name description: anyOf: - type: string - type: 'null' title: Description input_schema: anyOf: - additionalProperties: true type: object - type: 'null' title: Input Schema type: object required: - name title: AnthropicTool description: |- A tool definition. ``input_schema`` is Anthropic's name for what OpenAI calls ``function.parameters``. ValidationError: properties: loc: items: anyOf: - type: string - type: integer type: array title: Location msg: type: string title: Message type: type: string title: Error Type input: title: Input ctx: type: object title: Context type: object required: - loc - msg - type title: ValidationError ImageBlock: properties: type: type: string const: image title: Type default: image source: $ref: '#/components/schemas/ImageSource' cache_control: anyOf: - additionalProperties: true type: object - type: 'null' title: Cache Control type: object required: - source title: ImageBlock ToolUseBlock: properties: type: type: string const: tool_use title: Type default: tool_use id: type: string title: Id name: type: string title: Name input: additionalProperties: true type: object title: Input cache_control: anyOf: - additionalProperties: true type: object - type: 'null' title: Cache Control type: object required: - id - name title: ToolUseBlock description: >- An assistant's tool invocation. ``input`` is a decoded object here, while OpenAI carries it as a JSON *string* in ``function.arguments``. ToolResultBlock: properties: type: type: string const: tool_result title: Type default: tool_result tool_use_id: type: string title: Tool Use Id content: anyOf: - type: string - items: additionalProperties: true type: object type: array - type: 'null' title: Content is_error: type: boolean title: Is Error default: false cache_control: anyOf: - additionalProperties: true type: object - type: 'null' title: Cache Control type: object required: - tool_use_id title: ToolResultBlock description: >- The caller's answer to a ``tool_use``. Structurally this is the biggest difference from OpenAI: Anthropic puts tool results in a **user** message as content blocks, while OpenAI uses dedicated ``role="tool"`` messages. One Anthropic user turn therefore expands into several OpenAI messages — see ``messages_request_to_chat``. ImageSource: properties: type: type: string enum: - base64 - url title: Type media_type: anyOf: - type: string - type: 'null' title: Media Type data: anyOf: - type: string - type: 'null' title: Data url: anyOf: - type: string - type: 'null' title: Url type: object required: - type title: ImageSource description: >- Anthropic carries images as a typed source object rather than a URL string. ``base64`` names its own ``media_type``; ``url`` is a plain remote reference. Both invert to an OpenAI ``image_url`` — see ``_image_block_to_url``, the inverse of ``_image_source`` (``app/providers/anthropic.py:100-123``). securitySchemes: BearerAuth: type: http scheme: bearer description: >- Enter your MeshAPI key (`rsk_...`) — sent as `Authorization: Bearer `. ````