Anthropic Messages API & Claude Code

Endpoint & Authentication

The endpoint accepts your DEVUP AI API key in either of two standard header formats:

The anthropic-version: 2023-06-01 header is accepted and ignored — send it or omit it, the response is identical.

Verified Feature Support

Basic Request

Construct a standard Messages API payload specifying the target model, max token budget, and message history:

A final assistant message is kept as conversation history, and the model answers with a new assistant message.

curl https://api.devupai.com/anthropic/v1/messages \
  -H "Content-Type: application/json" \
  -H "x-api-key: $DEVUP_API_KEY" \
  -d '{
    "model": "claude-opus-5",
    "max_tokens": 1024,
    "messages": [
      {
        "role": "user",
        "content": "What is the capital of Algeria?"
      }
    ]
  }'

Response Shape

The API returns a standard Anthropic message structure:

{
  "id": "msg_01X9b8Cd3Ef7Gh2Jk4Lm5Np6",
  "type": "message",
  "role": "assistant",
  "model": "claude-opus-5",
  "content": [
    {
      "type": "text",
      "text": "The capital of Algeria is Algiers."
    }
  ],
  "stop_reason": "end_turn",
  "stop_sequence": null,
  "usage": {
    "input_tokens": 16,
    "output_tokens": 8
  }
}

Streaming

Pass stream: true to receive a Server-Sent Events (SSE) stream.

curl https://api.devupai.com/anthropic/v1/messages \
  -H "Content-Type: application/json" \
  -H "x-api-key: $DEVUP_API_KEY" \
  -d '{
    "model": "claude-opus-5",
    "max_tokens": 1024,
    "stream": true,
    "messages": [
      {
        "role": "user",
        "content": "Count from 1 to 3."
      }
    ]
  }'

Event Lifecycle

A streaming response delivers events in the following sequential order:

1. message_start — Initial message envelope with message ID and model metadata.
2. content_block_start — Starts a new content block index.
3. content_block_delta — Incremental text deltas (text_delta).
4. content_block_stop — Closes the active content block index.
5. message_delta — Final stop reason and usage statistics.
6. message_stop — Stream completion marker.

Complete wire format of a streaming exchange:

event: message_start
data: {"type":"message_start","message":{"id":"msg_01X9b8Cd3Ef7Gh2Jk4Lm5Np6","type":"message","role":"assistant","model":"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":"1, 2, 3."}}

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},"usage":{"input_tokens":14,"output_tokens":6}}

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

Tool Use

Supply a tools array defining JSON Schema parameters. When the model invokes a tool, the response contains a tool_use block and sets stop_reason: "tool_use".

Request with Tools

{
  "model": "claude-opus-5",
  "max_tokens": 1024,
  "tools": [
    {
      "name": "get_weather",
      "description": "Get the current weather for a location.",
      "input_schema": {
        "type": "object",
        "properties": {
          "location": {
            "type": "string",
            "description": "The city and country, e.g. Algiers, Algeria"
          }
        },
        "required": ["location"]
      }
    }
  ],
  "messages": [
    {
      "role": "user",
      "content": "What is the weather in Algiers?"
    }
  ]
}

Tool Call Response

{
  "id": "msg_01Y8kLm3N2P4Qr5St6Uv7Wx8",
  "type": "message",
  "role": "assistant",
  "model": "claude-opus-5",
  "content": [
    {
      "type": "tool_use",
      "id": "call_9a8b7c6d",
      "name": "get_weather",
      "input": {
        "location": "Algiers, Algeria"
      }
    }
  ],
  "stop_reason": "tool_use",
  "stop_sequence": null,
  "usage": {
    "input_tokens": 78,
    "output_tokens": 22
  }
}

For streaming requests with tool execution, arguments stream incrementally via content_block_delta events carrying input_json_delta payloads with partial_json fragments.

Models

The endpoint is model-agnostic: it accepts any model identifier from the DEVUP catalogue, not only claude-* names. Pass the model name exactly as listed in the catalogue:

Verified Models

verified 2026-08-26. Models not listed are expected to work but have not been tested on this endpoint.

Agent Requirement

Claude Code is an agent and requires a model that supports tool calling. A model without tool support will return text and the agent will not complete its loop.

{
  "type": "error",
  "error": {
    "type": "not_found_error",
    "message": "The requested model was not found."
  }
}

Limits

The largest accepted max_tokens is the model's own max_output_tokens in GET /v1/models, which differs from model to model. Read it for one model with GET /v1/models/deepseek-ai/DeepSeek-V4-Pro, for example.

• A max_tokens up to the model's max_output_tokens returns HTTP 200.
• A larger value returns HTTP 400 with an invalid_request_error whose message gives the model's limit, for example, for a model whose max_output_tokens is N:

{
  "type": "error",
  "error": {
    "type": "invalid_request_error",
    "message": "max_tokens must be between 1 and N."
  }
}

Model at Capacity

When the model is momentarily at capacity, the endpoint returns HTTP 529 with an overloaded_error and the Retry-After: 5 and retry-after-ms: 5000 headers. Retry after the indicated delay; the Anthropic SDKs retry it automatically.

{
  "type": "error",
  "error": {
    "type": "overloaded_error",
    "message": "This model is currently at capacity. Retry in a few seconds."
  }
}

Clients like Claude Code send their own max_tokens with each request. Set it to at most the model's max_output_tokens: for Claude Code, with CLAUDE_CODE_MAX_OUTPUT_TOKENS, as in the next section.

Claude Code

Configure Claude Code to route requests to DEVUP AI by setting environment variables in your active shell:

export ANTHROPIC_BASE_URL="https://api.devupai.com/anthropic"
export ANTHROPIC_API_KEY="sk-devup-..."
export ANTHROPIC_MODEL="deepseek-ai/DeepSeek-V4-Pro"
export CLAUDE_CODE_MAX_OUTPUT_TOKENS="8192"  # at most the model's max_output_tokens in GET /v1/models

Execution Modes

Run Claude Code with the configured environment:

claude --bare                 # interactive session
claude -p "your prompt"       # single prompt, non-interactive

Unrecognized Model Notice

Every invocation prints:

[claude-code:unrecognized_model] {"model":"...","query_source":"generate_session_title"}

This is emitted by the client, not by DEVUP AI, and is expected when the model name is not one the client ships with.

The --bare Requirement

Tested on Claude Code v2.1.238:

Tested on Claude Code v2.1.246:

The login behaviour changed between these two client versions, with both results measured on 2026-08-26, and the --bare guidance below still applies to clients that require it.

Claude Code documentation states that under --bare, authentication is strictly resolved via ANTHROPIC_API_KEY or apiKeyHelper, bypassing OAuth and system keychain lookups. Default interactive mode checks OAuth and the keychain first, prompting for a login rather than reading the environment variable.

Troubleshooting

If Claude Code was previously used in the same directory with another account, cached credentials in the local .claude folder take precedence over environment variables. Remove the local .claude directory and retry.

Anthropic SDK

Initialize official Anthropic SDKs by pointing the base URL parameter to https://api.devupai.com/anthropic:

import os
import anthropic

client = anthropic.Anthropic(
    base_url="https://api.devupai.com/anthropic",
    api_key=os.environ["DEVUP_API_KEY"],
)

message = client.messages.create(
    model="claude-opus-5",
    max_tokens=1024,
    messages=[
        {"role": "user", "content": "Hello!"}
    ],
)

print(message.content[0].text)