Claude Code setup

1. Create an API key on the main site. Make sure its group includes the model you intend to call.
2. Set the environment variables in the terminal where you launch Claude Code:

export ANTHROPIC_BASE_URL="https://api.nekocloud.vip"
export ANTHROPIC_AUTH_TOKEN="sk-your-key"
claude

Verify and troubleshoot

Send a short message. A 401 usually means the key was not passed or is invalid; a model unavailable error calls for checking the selected key's group. For persistent settings, refer to the Claude Code configuration documentation; avoid placing your real key in screenshots or support tickets. The Anthropic Base URL is the domain root, not /v1.

Model selection and conflicts

Copy an exact model ID accessible to the key's group from Plaza. Do not assume every Claude variant is available. Use claude --model REPLACE_WITH_MODEL_ID and confirm the selected model in the client.

Check for another provider's existing ANTHROPIC_API_KEY, Base URL or launch script. Avoid mixing authentication sources. Back up settings first and compare one auth method in a separate terminal; do not delete existing login data indiscriminately.

Safe input and persistence

In Bash, hide key input to keep it out of command history:

export ANTHROPIC_BASE_URL="https://api.nekocloud.vip"
read -rsp 'API Key: ' ANTHROPIC_AUTH_TOKEN
printf '\n'
export ANTHROPIC_AUTH_TOKEN
claude --model REPLACE_WITH_MODEL_ID

Temporary variables affect this terminal and newly launched children only. For desktop apps, Remote SSH or editor integrations, check where the client runs. Relaunch the client after changing variables, not the host. Shell/client files store secrets in plaintext; restrict permissions and protect backups.

Verify a real request

Send a short message, confirm the reply, then match it in Usage. Tool calls require separate verification. For 401 read the error code, for 404 check paths, and for 429 reduce load. Check records before retrying a timeout or broken stream. See Troubleshooting.

Tool-chain verification

In an isolated project, ask the agent to read a non-sensitive file and summarize it. Verify the tool call, result return and final answer. Chat success does not prove tools, images or long contexts work. Keep approval controls enabled; tests may incur charges.