Claude Code

View as Markdown

Claude Code speaks the Anthropic format (/v1/messages). Vikey accepts that format, so you point Claude Code at Vikey and tell it which model ids to use.

Tested with Claude Code 2.1.323 on 13 November 2026.

Recommended config

Write this to ~/.claude/settings.json (Windows: %USERPROFILE%\.claude\settings.json). If the file already exists, merge the contents instead of overwriting it.

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://api.vikey.ai",
    "ANTHROPIC_AUTH_TOKEN": "YOUR_API_KEY",
    "CLAUDE_CODE_ATTRIBUTION_HEADER": "0",
    "CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY": "1"
  },
  "modelOverrides": {
    "claude-opus-5-5": "anthropic/claude-opus-5.5",
    "claude-sonnet-5-5": "anthropic/claude-sonnet-5.5",
    "claude-haiku-5-5": "anthropic/haiku-5.5"
  }
}

Then run claude as usual. /model sonnet, /model opus and /model haiku now use the Claude models on Vikey.

Why each line is there

Line	Purpose
`ANTHROPIC_BASE_URL`	Sends every request to Vikey. Without `/v1`, because Claude Code adds `/v1/messages` itself.
`ANTHROPIC_AUTH_TOKEN`	Your Vikey API key, sent as `Authorization: Bearer`. It applies right away without a confirmation and takes precedence over a claude.ai login.
`CLAUDE_CODE_ATTRIBUTION_HEADER`	`0` removes a changing attribution block from the system prompt, so repeated prompt parts can be billed at the cache price.
`CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY`	`1` makes `/model` list the models from Vikey’s `GET /v1/models`, so you can pick other models from the list.
`modelOverrides`	Claude Code’s built-in model names (`claude-sonnet-5-5`) differ from Vikey model ids (`anthropic/claude-sonnet-5.5`). This mapping makes the `opus`, `sonnet` and `haiku` aliases and background tasks (session titles, summaries) use the right Vikey model, and Claude Code still knows what the model can do.

Use Vikey Auto or non-Claude models

Claude Code can also run models other than Claude, such as GPT or DeepSeek, because Vikey translates the format. Add model to settings.json:

{
  "model": "vikey/agent-coding",
  "env": {
    "ANTHROPIC_BASE_URL": "https://api.vikey.ai",
    "ANTHROPIC_AUTH_TOKEN": "YOUR_API_KEY",
    "CLAUDE_CODE_ATTRIBUTION_HEADER": "0",
    "CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY": "1"
  },
  "modelOverrides": {
    "claude-opus-5-5": "anthropic/claude-opus-5.5",
    "claude-sonnet-5-5": "anthropic/claude-sonnet-5.5",
    "claude-haiku-5-5": "anthropic/haiku-5.5"
  }
}

Or for one session only: claude --model vikey/agent-gpt.

Common model ids:

Model id	What it is
`vikey/agent-coding`	Vikey Auto with Claude models: Haiku 5.5, Sonnet 5.5, Opus 5.5
`vikey/agent-gpt`	Vikey Auto with GPT models: GPT-6 Luna, GPT-5.6 Terra, GPT-6.1 Sol
`vikey/agent-mocin`	The cheapest Vikey Auto, DeepSeek V4.1 Flash models
`openai/gpt-6.1-sol`, `deepseek/deepseek-v4.1-flash` and others	One fixed model. Full list on the pricing page.

With a model id Claude Code does not know, it prints Claude Code doesn't recognize the model .... That is only a warning: in the version we tested, Claude Code assumes a 1 million token context window, which matches what Vikey lists for these models. For another limit, set CLAUDE_CODE_MAX_CONTEXT_TOKENS, for example "CLAUDE_CODE_MAX_CONTEXT_TOKENS": "200000" in the env block.

How Vikey Auto works and what it costs is explained in Vikey Auto.

Try it without changing files

To try it in one terminal only:

macOS / Linux
PowerShell

export ANTHROPIC_BASE_URL=https://api.vikey.ai
export ANTHROPIC_AUTH_TOKEN=YOUR_API_KEY
export ANTHROPIC_DEFAULT_HAIKU_MODEL=openai/gpt-6-luna
claude --model vikey/agent-coding

$env:ANTHROPIC_BASE_URL = "https://api.vikey.ai"
$env:ANTHROPIC_AUTH_TOKEN = "YOUR_API_KEY"
$env:ANTHROPIC_DEFAULT_HAIKU_MODEL = "openai/gpt-6-luna"
claude --model vikey/agent-coding

ANTHROPIC_DEFAULT_HAIKU_MODEL replaces modelOverrides for background tasks. Without one of them, Claude Code sends its built-in model name for session titles and summaries, and those requests fail with Model not found.

Test the connection

Test the key and base URL directly, without Claude Code:

curl -sS https://api.vikey.ai/v1/messages \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{"model": "openai/gpt-6-luna", "max_tokens": 32, "messages": [{"role": "user", "content": "ping"}]}'

A response with "content" means the key and base URL are right.

Run claude, send one message, then type /status. The Anthropic base URL line should show https://api.vikey.ai, and the Auth token line should name ANTHROPIC_AUTH_TOKEN.

VS Code

The Claude Code extension for VS Code reads variables from VS Code’s own settings. Open Preferences: Open User Settings (JSON) and add:

{
  "claudeCode.environmentVariables": [
    { "name": "ANTHROPIC_BASE_URL", "value": "https://api.vikey.ai" },
    { "name": "ANTHROPIC_AUTH_TOKEN", "value": "YOUR_API_KEY" }
  ]
}

modelOverrides is still read from ~/.claude/settings.json.

Common problems

What you see	Cause	Fix
`Model 'claude-...' not found`	Claude Code sends its built-in model name	Add `modelOverrides` as above, or set `ANTHROPIC_DEFAULT_HAIKU_MODEL`
Asked to log in although curl works	The key is not read before the login screen	Put `ANTHROPIC_AUTH_TOKEN` in `~/.claude/settings.json` or export it in the shell, not in project settings
Warning about two login sources (`auth may not work as expected`)	The Vikey key and a claude.ai login are both active	Run `/logout` so only the Vikey key is used
`ANTHROPIC_API_KEY` ignored	This variable needs a one-time approval in interactive mode	Use `ANTHROPIC_AUTH_TOKEN`, or approve it in `/config`
`400` with `Extra inputs are not permitted`	A Claude Code beta feature the target model does not take	Add `"CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS": "1"` to the `env` block
`402 Insufficient balance`	No balance left	Top up and continue the session