Claude Code

Claude Code is Anthropic's agentic coding CLI, which also runs inside VS Code and JetBrains. It speaks the Anthropic Messages API, so you point it at https://tokens.bd (no /v1) and it sends requests to /v1/messages.

Quick setup with the Tokens CLI#

The Tokens CLI signs you in through the browser, creates a key, lets you pick a model and writes Claude Code's settings for you. It needs Node 18 or newer.

curl -fsSL https://tokens.bd/cli/tokens.mjs -o tokens.mjs && node tokens.mjs setup --base-url https://tokens.bd --agents claude

iwr https://tokens.bd/cli/tokens.mjs -OutFile tokens.mjs; node tokens.mjs setup --base-url https://tokens.bd --agents claude

It merges three values into the env block of ~/.claude/settings.json: ANTHROPIC_BASE_URL, ANTHROPIC_AUTH_TOKEN and ANTHROPIC_MODEL. The old file is kept as settings.json.tokens-backup-<timestamp>. If the file has comments or a syntax error, the CLI leaves it alone and points you to /dashboard/connect instead.

The CLI does not set the alias and beta variables described below. Add them afterwards; they prevent most of the odd behavior people hit with non-Claude models.

Set ANTHROPIC_BASE_URL for Claude Code manually#

Install Claude Code first if you haven't (curl -fsSL https://claude.ai/install.sh | bash on macOS/Linux, irm https://claude.ai/install.ps1 | iex in PowerShell, or npm install -g @anthropic-ai/claude-code). Then create a key at /dashboard/keys (see API keys).

Option A: environment variables#

Good for trying things out, or if you never want the key in a file.

export TOKENS_API_KEY="tok_live_your_key"
export ANTHROPIC_BASE_URL="https://tokens.bd"
export ANTHROPIC_AUTH_TOKEN="$TOKENS_API_KEY"
export ANTHROPIC_MODEL="deepseek/deepseek-v4.1-flash"
export ANTHROPIC_DEFAULT_OPUS_MODEL="deepseek/deepseek-v4.1-flash"
export ANTHROPIC_DEFAULT_SONNET_MODEL="deepseek/deepseek-v4.1-flash"
export ANTHROPIC_DEFAULT_HAIKU_MODEL="deepseek/deepseek-v4.1-flash"
export CLAUDE_CODE_SUBAGENT_MODEL="deepseek/deepseek-v4.1-flash"
export CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS=1

$env:TOKENS_API_KEY = "tok_live_your_key"
$env:ANTHROPIC_BASE_URL = "https://tokens.bd"
$env:ANTHROPIC_AUTH_TOKEN = $env:TOKENS_API_KEY
$env:ANTHROPIC_MODEL = "deepseek/deepseek-v4.1-flash"
$env:ANTHROPIC_DEFAULT_OPUS_MODEL = "deepseek/deepseek-v4.1-flash"
$env:ANTHROPIC_DEFAULT_SONNET_MODEL = "deepseek/deepseek-v4.1-flash"
$env:ANTHROPIC_DEFAULT_HAIKU_MODEL = "deepseek/deepseek-v4.1-flash"
$env:CLAUDE_CODE_SUBAGENT_MODEL = "deepseek/deepseek-v4.1-flash"
$env:CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS = "1"

$env: lasts for the current PowerShell window. Use setx NAME "value" for each variable to keep it in new windows.

Option B: ~/.claude/settings.json#

Values in the settings file's env block win over shell exports. On Windows the file is %USERPROFILE%\.claude\settings.json.

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://tokens.bd",
    "ANTHROPIC_AUTH_TOKEN": "tok_live_your_key",
    "ANTHROPIC_MODEL": "deepseek/deepseek-v4.1-flash",
    "ANTHROPIC_DEFAULT_OPUS_MODEL": "deepseek/deepseek-v4.1-flash",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "deepseek/deepseek-v4.1-flash",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "deepseek/deepseek-v4.1-flash",
    "CLAUDE_CODE_SUBAGENT_MODEL": "deepseek/deepseek-v4.1-flash",
    "CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS": "1",
    "CLAUDE_CODE_MAX_CONTEXT_TOKENS": "128000",
    "CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC": "1"
  }
}

128000 is a placeholder. Set CLAUDE_CODE_MAX_CONTEXT_TOKENS to the real context window shown on the model's page in /models.

What the variables do:

VS Code extension#

Set the same variables in your VS Code user settings:

{
  "claudeCode.environmentVariables": [
    { "name": "ANTHROPIC_BASE_URL", "value": "https://tokens.bd" },
    { "name": "ANTHROPIC_AUTH_TOKEN", "value": "tok_live_your_key" },
    { "name": "ANTHROPIC_MODEL", "value": "deepseek/deepseek-v4.1-flash" }
  ]
}

Switch models#

Change ANTHROPIC_MODEL and the four alias variables to another id and restart Claude Code. Keep them in sync, or the opus alias and background tasks keep running the old model. Copy exact ids from /models, from node tokens.mjs models, or from GET /v1/models. Choosing a model covers which models suit agent work.

The /model picker won't list Tokens models on its own: Claude Code's gateway model discovery keeps only ids containing claude or anthropic. To add one extra entry to the picker, set ANTHROPIC_CUSTOM_MODEL_OPTION to a model id.

Verify it works#

Test the endpoint directly first (macOS, Linux, WSL or Git Bash):

curl -sS -w '\n%{http_code}\n' -X POST "https://tokens.bd/v1/messages" \
  -H "Authorization: Bearer $TOKENS_API_KEY" \
  -H "anthropic-version: 2023-06-01" -H "content-type: application/json" \
  -d '{"model": "deepseek/deepseek-v4.1-flash", "max_tokens": 1, "messages": [{"role": "user", "content": "."}]}'

A 200 means the key and model are fine. Then run claude, type /status, and check that the base URL line shows https://tokens.bd and the auth token line names ANTHROPIC_AUTH_TOKEN. claude doctor validates your settings files. The request should also appear in Usage analytics in the dashboard.

Troubleshooting#

404 on every request. The base URL ends in /v1. Claude Code appends /v1/messages itself, so use https://tokens.bd.

401 invalid_api_key even though you exported a new key. The env block in ~/.claude/settings.json overrides your shell. Check which value /status reports.

400 errors about thinking, effort or unknown fields. For model ids it doesn't recognize, Claude Code sends adaptive thinking, effort and beta tool fields. CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS=1 removes most of them, though not thinking or effort; Claude Code retries without those when the upstream rejects them. If a model still fails, try another one.

The session never compacts, or hits "prompt too long". Claude Code assumes 200K tokens of context for unknown ids. Set CLAUDE_CODE_MAX_CONTEXT_TOKENS to the model's real window. CLAUDE_CODE_AUTO_COMPACT_WINDOW (minimum 100,000) and CLAUDE_CODE_MAX_OUTPUT_TOKENS give finer control.

Unexpected model in your usage log. Background tasks use ANTHROPIC_DEFAULT_HAIKU_MODEL, or the main model if it's unset.

Features missing. Remote Control and voice dictation are disabled while a gateway credential is set, and fast mode checks Anthropic's API directly, so it won't work through Tokens.

403 model_not_allowed_on_key or tier_permission_denied, 402, or 429 window_exhausted. These are account limits, not Claude Code problems. See Troubleshooting and include the x-tokens-request-id header value if you open a ticket.