Claude Code Setup

Claude Code is Anthropic's terminal coding agent. It speaks the Anthropic Messages API -- which ClawRouter exposes on every provider -- so you can route Claude Code through ClawRouter to any configured provider, including non-Anthropic ones.

Version 1.0.18

Prerequisites

ClawRouter running -- clawrouter status (default: http://localhost:3030).
A provider created in the ClawRouter dashboard, with API keys added.
The proxy API key -- copy it from Settings > Proxy API Key in the dashboard.

Method 1: "Prompt for AI" (Recommended)

Open the provider's detail page in the ClawRouter dashboard.
Click "Prompt for AI" on the Base URL banner and select the Claude Code tab.
Click Copy and paste the prompt to your AI agent -- or follow the generated instructions yourself.

The generated prompt contains the exact env block below with the correct Base URL, your saved model IDs, and your real proxy API key.

Method 2: Manual Configuration

Claude Code reads its connection settings from environment variables. Add them to the env block of ~/.claude/settings.json (merge with any existing keys -- preserve everything else), then restart Claude Code:

json

{
  "env": {
    "ANTHROPIC_BASE_URL": "http://localhost:3030/proxy/my-provider-id",
    "ANTHROPIC_AUTH_TOKEN": "cr_your_proxy_key",
    "ANTHROPIC_MODEL": "model-id",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "model-id"
  }
}

Variable	Description
`ANTHROPIC_BASE_URL`	The provider's ClawRouter endpoint without a `/v1` suffix -- Claude Code appends `/v1/messages` itself. Do not add it
`ANTHROPIC_AUTH_TOKEN`	The proxy API key from Settings > Proxy API Key. Sent as `Authorization: Bearer ...` -- do not add the "Bearer " prefix yourself
`ANTHROPIC_MODEL`	The model ID to use. Get model IDs from the provider's Models tab > Fetch Models
`ANTHROPIC_DEFAULT_HAIKU_MODEL`	The model used for background/small tasks -- set it to a model that exists on the same provider

Works with every provider: for non-Anthropic providers (OpenAI, Google Gemini, Groq, ...), ClawRouter translates the Anthropic Messages format automatically and injects the Claude Code identity/system preamble, so the experience matches native Anthropic behavior. The Kimi for Coding and MiniMax Coding presets speak Anthropic Messages natively -- no translation needed.

Optional: Model Parameters & Gateway Discovery

Add these to the same env block when useful (merge, don't replace):

Variable	Description
`CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY`	Set to `"1"` -- Claude Code populates its `/model` picker from the gateway's `/v1/models` (ClawRouter proxies it upstream), so you can switch models without editing the config
`CLAUDE_CODE_MAX_CONTEXT_TOKENS`	The model's real context window -- Claude Code may not know it for custom/gateway model IDs. Look up the exact value in the provider's model docs
`CLAUDE_CODE_MAX_OUTPUT_TOKENS`	Output cap -- Claude Code assumes 32000 for unrecognized model IDs; set the model's real max output
`CLAUDE_CODE_EFFORT_LEVEL`	Reasoning effort: `low` / `medium` / `high` / `xhigh` / `max` (in-session: `/effort`)
`MAX_THINKING_TOKENS`	Fixed extended-thinking budget in tokens; `0` disables thinking on third-party providers

Reasoning effort value sets differ per model (one takes low/medium/high, another low/medium/xhigh, another low/high/max) -- always verify in the provider's official model docs. ClawRouter translates thinking/reasoning parameters into the provider's native dialect automatically.

Full reference: https://code.claude.com/docs/en/env-vars (machine-readable index: https://code.claude.com/docs/llms.txt; every docs page also serves markdown at <page>.md).

Choosing the Provider & Model

Which upstream provider answers is decided by ANTHROPIC_BASE_URL -- each ClawRouter provider has its own /proxy/{provider-id} endpoint. Point the variable at a different provider ID to switch providers.
Which model answers is decided by ANTHROPIC_MODEL (and ANTHROPIC_DEFAULT_HAIKU_MODEL for small tasks). ClawRouter forwards the model ID upstream as-is; with Model Fallback enabled, a failing model is automatically substituted.

Verify the Connection

Restart Claude Code after editing settings.json.
Run /status -- it should show the ClawRouter base URL.
Send a short prompt and check the ClawRouter Logs page for the successful request.

You can also verify the endpoint directly with curl (Anthropic Messages style):

bash

curl -X POST http://localhost:3030/proxy/my-provider-id/v1/messages \
  -H "Content-Type: application/json" \
  -H "x-api-key: cr_your_proxy_key" \
  -H "anthropic-version: 2023-06-01" \
  -d '{ "model": "model-id", "max_tokens": 50, "messages": [{"role": "user", "content": "Hello"}] }'

Troubleshooting

HTTP 401 "Invalid or missing API key"

ANTHROPIC_AUTH_TOKEN is missing, wrong, or was regenerated. Copy the current key from Settings > Proxy API Key. This 401 comes from ClawRouter itself, before any upstream request.
Make sure you did not add a "Bearer " prefix -- Claude Code adds it automatically.

Connection refused

Ensure ClawRouter is running: clawrouter status.
Verify ANTHROPIC_BASE_URL points at the right host/port (default: 3030) and does not end with /v1 -- Claude Code appends /v1/messages itself; a /v1 in the base URL produces a broken path.

"Model not found" errors

The model ID must exist on the upstream provider. Use Fetch Models in the provider's Models tab for the current list.
Enable Model Fallback with backup models so a stale ID fails over automatically.

Changes not taking effect

Claude Code reads settings.json at startup -- restart it after every edit.