POST /v1/messages: Messages

POST /v1/messages takes Anthropic's request shape (system, messages of content blocks, a required max_tokens, tools with input_schema) and answers in Anthropic's message object or its named-event stream. Behind it is the same pipeline as every other wire: the same routing, memory, connector tools, agent mode and allowances.

It exists so Claude Code and the Anthropic SDKs can use Memtro as their model by setting ANTHROPIC_BASE_URL to https://platform.memtro.com. Errors come back in Anthropic's shape, with a type an Anthropic client has a case for.

Example

Against https://platform.memtro.com, with your key in MEMTRO_API_KEY.

curl https://platform.memtro.com/v1/messages \ -H "x-api-key: $MEMTRO_API_KEY" \ -H "content-type: application/json" \ -d '{ "model": "auto", "max_tokens": 1024, "system": "Be terse.", "messages": [{"role": "user", "content": "What changed on the Baltic account this week?"}] }'

from anthropic import Anthropic client = Anthropic(base_url="https://platform.memtro.com", api_key="sk-memtro-…") message = client.messages.create( model="auto", max_tokens=1024, system="Be terse.", messages=[{"role": "user", "content": "What changed on the Baltic account this week?"}], ) print(message.content[0].text)

import Anthropic from "@anthropic-ai/sdk"; const client = new Anthropic({ baseURL: "https://platform.memtro.com", apiKey: process.env.MEMTRO_API_KEY }); const message = await client.messages.create({ model: "auto", max_tokens: 1024, system: "Be terse.", messages: [{ role: "user", content: "What changed on the Baltic account this week?" }], }); console.log(message.content);

A non-streaming reply.

{
  "id": "msg-2fJw…",
  "type": "message",
  "role": "assistant",
  "model": "anthropic/claude-opus-5",
  "content": [{ "type": "text", "text": "Two renewals and a price change." }],
  "stop_reason": "end_turn",
  "stop_sequence": null,
  "usage": { "input_tokens": 1830, "output_tokens": 11, "cache_read_input_tokens": 2048, "cache_creation_input_tokens": 0 },
  "memtro": { "mode": "chat", "routing": { "requested": "auto", "model": "anthropic/claude-opus-5", "complexity": 2, "method": "heuristic", "reason": "…", "fallbacks": ["…"], "kind": "general" } }
}

Authentication

An API key, as x-api-key (what the Anthropic SDKs send) or Authorization: Bearer.

Every reply, errors included, carries x-request-id and request-id headers with the same id, starting msg-. Quote it to support.
Browsers may call it from any origin: it answers CORS preflights and allows the Authorization, x-api-key and Anthropic version headers.

Request body

JSON, with Content-Type: application/json.

modelstringRequired

A model id or routing id; see Model ids. This wire has no default.

messagesarray of objectsRequired

The conversation.

2 fields of messages

rolestringRequired: user or assistant. system and developer turns are moved into the system prompt, and any other role is read as user.
contentstring or array of objectsRequired: Text, or content blocks: text; image (source with type base64 and media_type + data, or type url and url); tool_use (id, name, input); tool_result (tool_use_id, content, is_error). A block may carry cache_control. Unknown block types are dropped; an image URL that cannot be used is a 400.

max_tokensintegerRequired

A ceiling on the answer's tokens; a positive integer.

systemstring or array of objectsOptional

The system prompt, as text or text blocks.

toolsarray of objectsOptional

Your own tools, {name, description, input_schema}, returned as tool_use blocks for you to run. Anthropic's server tools (any other type than none or custom) are left out without an error.

tool_choiceobjectOptional

{"type": "auto" | "any" | "tool" | "none"}, with name for tool.

streambooleanOptional

Answer as Anthropic's named-event stream. See Streaming.

Default:

false

temperaturenumberOptional

Passed to the provider.

top_pnumberOptional

Passed to the provider.

top_knumberOptional

Accepted and not forwarded.

stop_sequencesarray of stringsOptional

Where to stop generating.

metadataobjectOptional

user_id is accepted and ignored.

thinkingobjectOptional

Anthropic's thinking request (type, budget_tokens, display): asks for the thinking back and how hard to think.

output_configobjectOptional

effort, as current Claude Code sends it, is read as reasoning_effort.

memtroobjectOptional

Memtro's options; see the memtro object. {"memtro": {"mode": "agent"}} gets a multi-step run.

Response

200application/json

An Anthropic message.

idstringOptional

msg-…, the same id as the request-id header.

type"message"Optional

Always message.

role"assistant"Optional

Always assistant.

modelstringOptional

The model that was picked.

contentarray of objectsOptional

thinking (when asked for), text, and a tool_use block for each call to your tools. Memtro's own tools ran on the server; their results are in the answer.

stop_reasonstringOptional

Why it ended.

One of end_turn, max_tokens, tool_use or refusal.

stop_sequencenullOptional

Always null.

usageobjectOptional

input_tokens, output_tokens, cache_read_input_tokens, cache_creation_input_tokens.

memtroobjectOptional

As on chat completions, without answered_by; see What comes back. On a stream it rides on message_delta.

200text/event-stream

With stream: true: Anthropic's named events. See Streaming.

Streaming

Anthropic's event stream: message_start, then for each block a content_block_start, its content_block_delta events and a content_block_stop, then message_delta with the stop_reason and message_stop. Blocks are thinking (when asked for), text, and tool_use for calls to your tools. If the run fails after the stream has started, an error event with type api_error is sent instead and no message_stop follows.

event: message_start
data: {"type":"message_start","message":{"id":"msg-2fJw…","type":"message","role":"assistant","model":"anthropic/claude-opus-5","content":[],"stop_reason":null,"stop_sequence":null,"usage":{"input_tokens":0,"output_tokens":0}}}

event: content_block_start
data: {"type":"content_block_start","index":0,"content_block":{"type":"text","text":""}}

event: content_block_delta
data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":"Two renewals and a price change."}}

event: content_block_stop
data: {"type":"content_block_stop","index":0}

event: message_delta
data: {"type":"message_delta","delta":{"stop_reason":"end_turn","stop_sequence":null},…}

event: message_stop
data: {"type":"message_stop"}

Errors

In Anthropic's envelope, {"type": "error", "error": {"type", "message"}}, with no code: the type is chosen from the status. See Errors on the overview.

Status	Type	When
400	`invalid_request_error`	The body failed validation, or another 400 from the error catalogue, such as a refused OpenAI-compatible endpoint; the message says which.
401	`authentication_error`	The key is missing, malformed, revoked or expired.
402	`subscription_inactive`	The organisation has no active subscription.
404	`not_found_error`	Unknown model id.
429	`rate_limit_error`	The month's allowance for the request's level is spent.
500 and above	`api_error`	The provider or Memtro failed; `overloaded_error` for a 529.

Notes

Anthropic's error body has no code: the type is chosen from the status, as in the table under Errors. A 429 on this wire is an allowance, not a rate limit, whatever the type says.

Messages