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

# Anthropic Messages API

> Call models labelled anthropic with the native Anthropic Messages protocol.

Use this API for models labelled `anthropic` in the Model Marketplace or the model list for your current key. Read the [API and model capability matrix](/en/developer/capability-matrix) first: **the model, key group, and protocol must all match**.

<Warning>
  **Messages is not a universal endpoint**

  Use [Chat Completions](/en/developer/openai-compatible) for OpenAI-compatible applications. Do not switch a current Anthropic-type Claude channel to `/v1/responses`; `not implemented` normally means a protocol mismatch, not a temporary error that retries can solve.
</Warning>

## Endpoint and minimum request

```text theme={"system"}
POST https://nexusapi.link/v1/messages
```

```bash theme={"system"}
curl 'https://nexusapi.link/v1/messages' \
  -H 'Content-Type: application/json' \
  -H 'Authorization: Bearer YOUR_API_KEY' \
  -H 'anthropic-version: 2023-06-01' \
  -d '{
    "model": "YOUR_ANTHROPIC_MODEL_ID",
    "max_tokens": 256,
    "messages": [{"role": "user", "content": "Hello"}]
  }'
```

| Item                | Required                        | Notes                                                              |
| ------------------- | ------------------------------- | ------------------------------------------------------------------ |
| `model`             | Yes                             | Exact model ID visible to the current key and labelled `anthropic` |
| `max_tokens`        | Yes                             | Maximum output tokens for this request                             |
| `messages`          | Yes                             | Conversation in Anthropic Messages format                          |
| `anthropic-version` | Required by the native protocol | Managed by the SDK; use a compatible version for manual calls      |

Use the [Apifox API reference](https://nexusapi.apifox.cn/) and the protocol documentation for your Anthropic client for full fields, tools, streaming, and response schemas. Optional-field availability still depends on the selected model and channel.

## Use the Anthropic SDK

The native SDK manages its required protocol headers. Pass a NexusAPI key to the SDK and set its Base URL **without `/v1`**:

```python theme={"system"}
from anthropic import Anthropic

client = Anthropic(
    api_key="YOUR_API_KEY",
    base_url="https://nexusapi.link",
)

message = client.messages.create(
    model="YOUR_ANTHROPIC_MODEL_ID",
    max_tokens=256,
    messages=[{"role": "user", "content": "Hello"}],
)
```

The SDK appends `/v1/messages`. Do not set `base_url` to `https://nexusapi.link/v1/messages`, which can create a repeated path or 404.

## Claude Code versus your own app

| Scenario                   | Configuration                                                                                                                                                 |
| -------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Claude Code                | Follow the [Claude Code guide](/en/tools/claude-code): set `ANTHROPIC_BASE_URL=https://nexusapi.link` and `ANTHROPIC_AUTH_TOKEN` in `~/.claude/settings.json` |
| Anthropic SDK / native app | Set Base URL to `https://nexusapi.link` and let the client send Messages                                                                                      |
| OpenAI SDK app             | Do not mix this setup into it; use [Chat Completions](/en/developer/openai-compatible)                                                                        |

## Troubleshooting

* `401`: Check that the key is correct, enabled, and has not been reset after exposure.
* `404`: Check whether an SDK has appended a second `/v1`.
* `400`: Verify with only `model`, `max_tokens`, and `messages`, then add optional fields gradually.
* `500, not implemented`: Make sure an Anthropic-type model is not being sent through `/v1/responses`.

For persistent errors, keep the Request ID and a redacted request, then use the [self-check guide](/en/faq/self-check).