Use your Claude subscription with Claude Code

Send Claude Code requests through LiteLLM using your own Claude Pro, Max, Team, or Enterprise subscription. LiteLLM applies access controls, budgets, and guardrails and records usage. Claude Code keeps your login and refreshes it.

Prerequisites

• Claude Code installed and a Claude subscription. Team and Enterprise members use the account their administrator invited.
• A running LiteLLM proxy Since v1.81.14 with PostgreSQL for virtual keys and logs.

Step 1: Configure the proxy

Add this entry to your proxy's model_list and restart the proxy. Skip this step if your administrator has already configured it. Preserve your existing master key, database, and other settings.

config.yaml

model_list:
  - model_name: "anthropic/*"
    litellm_params:
      model: "anthropic/*"

The wildcard routes whichever Claude model your plan allows. Keep this deployment configured as shown to use each developer's subscription login. LiteLLM forwards that login automatically.

Step 2: Create a virtual key

Open your proxy dashboard at http://localhost:4000/ui (replace the URL with your proxy's). For a local proxy, sign in as admin with your master key. Go to Virtual Keys, click + Create New Key, enter a Key Name, and select All anthropic models under Models. Click Create Key and copy the key; it is shown only once.

For an existing key, allow anthropic/*. Leaving Models empty allows every model on the proxy.

Step 3: Connect Claude Code

If you previously used API-key authentication, remove apiKeyHelper and the ANTHROPIC_AUTH_TOKEN / ANTHROPIC_API_KEY entries in env from your active Claude Code settings: ~/.claude/settings.json and your project's .claude/settings.json or .claude/settings.local.json.

From your project folder, clear shell overrides, set your proxy's root URL and virtual key, and launch Claude Code:

unset ANTHROPIC_AUTH_TOKEN ANTHROPIC_API_KEY ANTHROPIC_MODEL
export ANTHROPIC_BASE_URL="http://localhost:4000"
export ANTHROPIC_CUSTOM_HEADERS="x-litellm-api-key: Bearer sk-<your-litellm-virtual-key>"

claude

Replace the URL with your proxy's root address, such as https://litellm.example.com, and replace the key. Retain any existing custom headers on separate lines.

Choose Claude account with subscription and complete browser sign-in if needed. If the browser does not open, press c to copy the URL; paste a code back into the terminal if prompted. Complete the workspace trust prompt for your project. An existing subscription login can stay signed in.

Use /model to select a Claude model your subscription supports. To set a default explicitly, set ANTHROPIC_MODEL to a Claude model ID such as claude-opus-5. For future terminals, add the unset and export lines to the end of your shell profile (~/.zshrc or ~/.bashrc).

Step 4: Verify a request

Send: Reply with exactly Hello.

You should receive a greeting. In the dashboard, open Logs and find the successful session under your virtual key. Open it to inspect the model, tokens, and cost.

Confirm subscription usage

Filter Logs with Credential: Client OAuth token Since v1.105.0; Request Details shows the same value. Spend-log metadata records used_client_oauth_token=true for subscription requests and false for configured credentials. For API access, query /spend/logs/ui with that filter and UTC start_date / end_date values in YYYY-MM-DD HH:MM:SS format.

Limits and billing

Your Claude plan controls model access and quota. Use the LiteLLM virtual key's user and team for gateway attribution, and Claude's admin console for subscription membership.

LiteLLM reports cost at Anthropic API list prices, even for subscription requests. Use these amounts for budgets in API-equivalent dollars. Anthropic bills subscription requests through your plan; calculate API-billed spend from rows that used configured API credentials.

Have each developer sign in with their own subscription, following Anthropic's usage terms. Team and Enterprise administration stays in Claude's admin console; use forceLoginOrgUUID to restrict Claude Code to your organization.

Credentials and supported routes

How it works

x-litellm-api-key authenticates to LiteLLM. Claude Code supplies its subscription OAuth token in Authorization; LiteLLM forwards it to direct anthropic/ deployments, ahead of any configured API key, and adds the required OAuth beta header. Other client headers require separate header forwarding configuration.

Which requests use the subscription token

Use direct Anthropic deployments on /v1/messages or /v1/chat/completions for subscription forwarding and credential attribution. Bedrock and Vertex AI use their own credentials (provider isolation works Since v1.99.0). /v1/messages/count_tokens uses configured credentials. The /anthropic pass-through route uses the proxy's Anthropic key when configured; use the proxy root for the subscription setup in this guide.

Credential storage, refresh, and revocation

Claude Code stores and refreshes its login locally. LiteLLM forwards each caller's token for that request and records the credential type in spend logs. Use /logout to sign out or /login to renew an expired or revoked login.

To end gateway access, block or delete the LiteLLM key. To end subscription access too, sign out of Claude Code or ask the Claude administrator to revoke it.

Troubleshooting

Problem	Fix
LiteLLM rejects the key	Check the URL and x-litellm-api-key in ANTHROPIC_CUSTOM_HEADERS. A deleted or incorrect key returns 401.
Anthropic rejects the login, or requests use an API key	Check /status, remove API credential overrides and apiKeyHelper, then run /login. Confirm a direct anthropic/ deployment and the proxy root URL.
Model not found or access denied	Configure anthropic/*, allow All anthropic models on the key, and choose a model your plan supports.
No request appears in LiteLLM	Check ANTHROPIC_BASE_URL, then restart Claude Code from the terminal where you exported it.
Healthy accounts receive 429 No deployments available for selected model	Ask the operator to check for an existing deployment cooldown; follow the recovery steps.

Check gateway authentication and model access with your URL and key:

curl "http://localhost:4000/v1/models" \
  -H "x-litellm-api-key: Bearer sk-<your-litellm-virtual-key>"

A 200 confirms gateway authentication and lists accessible models. For a 401, replace the key with an active virtual key. Use the prompt and Credential filter in Step 4 to confirm subscription access.

See the setup video, subscription options, or Claude Code with Lite CLI SSO.