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

# Claude Code setup guide

> Connect Claude Code to the powertokens platform through the Anthropic Messages-compatible path.

Claude Code uses the Anthropic Messages protocol. To use it with the `powertokens platform`, point Claude Code at `https://api.powertokens.ai` and choose a text model available through the Anthropic Messages path. The recommended platform default is `glm-5.2`; the recommended MiniMax coding-agent path is `MiniMax-M3`.

If you need to choose the endpoint first, see [Text model protocols and endpoints](/en/ecosystem-tools/text-model-protocols).

## Available models

Use the following model IDs with Claude Code. For coding-agent workflows, prefer `glm-5.2` or `MiniMax-M3`.

| Upstream provider | Model ID                 | Recommended use                                                             |
| ----------------- | ------------------------ | --------------------------------------------------------------------------- |
| Zhipu             | `glm-4.5-air`            | Lightweight coding help                                                     |
| Zhipu             | `glm-4.7`                | General development tasks                                                   |
| Zhipu             | `glm-4.7-flash`          | Fast lightweight coding tasks                                               |
| Zhipu             | `glm-5`                  | Deeper reasoning and edits                                                  |
| Zhipu             | `glm-5-turbo`            | Faster coding tasks                                                         |
| Zhipu             | `glm-5.2`                | Heavier reasoning; configure manually if it is not shown in model discovery |
| MiniMax           | `MiniMax-M2.5`           | General agentic coding                                                      |
| MiniMax           | `MiniMax-M2.5-highspeed` | Latency-sensitive tasks                                                     |
| MiniMax           | `MiniMax-M2.7`           | General development and longer tasks                                        |
| MiniMax           | `MiniMax-M2.7-highspeed` | Faster long tasks                                                           |
| MiniMax           | `MiniMax-M3`             | Complex coding-agent work                                                   |
| Alibaba Qwen      | `qwen3-coder-plus`       | Optional coding model                                                       |
| Alibaba Qwen      | `qwen3-max`              | Optional general reasoning model                                            |
| Alibaba Qwen      | `qwen3.5-flash`          | Optional lightweight model                                                  |
| Alibaba Qwen      | `qwen3.6-plus`           | Optional balanced model                                                     |
| BytePlus DeepSeek | `deepseek-v3-2-251201`   | Optional text model; passed opencode Anthropic-provider tool validation     |
| BytePlus Seed     | `seed-1-6-250915`        | Optional text model                                                         |
| BytePlus Seed     | `seed-1-8-251228`        | Optional text model                                                         |
| BytePlus Seed     | `seed-2-0-lite-260228`   | Optional lightweight model                                                  |
| BytePlus Seed     | `seed-2-0-mini-260215`   | Optional lightweight model                                                  |
| BytePlus Seed     | `seed-2-0-pro-260328`    | Optional heavier text model                                                 |

## Prepare your environment

Install or update Claude Code:

```bash theme={null}
npm install -g @anthropic-ai/claude-code
claude --version
```

Clear old Anthropic environment variables before switching providers:

```bash theme={null}
unset ANTHROPIC_BASE_URL
unset ANTHROPIC_API_KEY
unset ANTHROPIC_AUTH_TOKEN
unset ANTHROPIC_MODEL
```

## Configure Claude Code

Add this to user-level `~/.claude/settings.json` or project-level `.claude/settings.json`:

```json theme={null}
{
  "env": {
    "ANTHROPIC_BASE_URL": "https://api.powertokens.ai",
    "ANTHROPIC_AUTH_TOKEN": "YOUR_POWERTOKENS_API_KEY",
    "ANTHROPIC_MODEL": "glm-5.2",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "glm-5-turbo",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "glm-5.2",
    "ANTHROPIC_DEFAULT_OPUS_MODEL": "MiniMax-M3",
    "CLAUDE_CODE_SUBAGENT_MODEL": "glm-5.2",
    "CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY": "1"
  }
}
```

Claude Code should use `ANTHROPIC_AUTH_TOKEN` so the client sends a bearer token. Do not set both `ANTHROPIC_AUTH_TOKEN` and `ANTHROPIC_API_KEY`; Claude Code gives `ANTHROPIC_AUTH_TOKEN` precedence. If your runtime explicitly requires `x-api-key` auth, replace `ANTHROPIC_AUTH_TOKEN` with `ANTHROPIC_API_KEY`.

## Start and verify

Run Claude Code from your project:

```bash theme={null}
claude
```

Inside the session, check:

```text theme={null}
/status
/model
```

If the model picker does not show platform models, set `ANTHROPIC_MODEL` manually to one of the IDs above. Some discovery flows only show model IDs that begin with `claude` or `anthropic`; manual model IDs still work when the gateway accepts them.

Verify that Claude Code reads the configuration in non-interactive mode:

```bash theme={null}
claude -p --model glm-5.2 "Reply exactly: ok"
```

Then verify tool use with a local file:

```bash theme={null}
printf "powertokens-compat-ok\n" > compat.txt
claude -p --model glm-5.2 --tools=Read --allowedTools=Read \
  "Use the Read tool to read compat.txt. Reply exactly with the file contents."
rm compat.txt
```

You can also verify the same model directly through the platform API.

```bash theme={null}
curl -X POST "https://api.powertokens.ai/v1/messages" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_POWERTOKENS_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  --data '{
    "model": "glm-5.2",
    "max_tokens": 64,
    "messages": [
      {"role": "user", "content": "Reply with ok."}
    ],
    "stream": false
  }'
```

## Model choice

| Scenario                  | Suggested model                                                    |
| ------------------------- | ------------------------------------------------------------------ |
| Default coding agent      | `glm-5.2`                                                          |
| Small edits               | `glm-5-turbo`                                                      |
| Complex code review       | `glm-5.2` or `MiniMax-M3`                                          |
| Parallel subagents        | `glm-5.2` as main model and `glm-5-turbo` as the lightweight model |
| MiniMax coding-agent path | `MiniMax-M3`                                                       |

## Not default recommendations

The following models may declare `anthropic` in model discovery, but current ecosystem-tool validation does not recommend them as Claude Code defaults:

| Model                   | Reason                                                                                   |
| ----------------------- | ---------------------------------------------------------------------------------------- |
| `seed-1-6-flash-250715` | The upstream provider does not support `/v1/messages`; use an OpenAI-compatible provider |