Access & base URLs
Choose the connection your application needs. The model API, MCP tools and dashboard share your account balance, but use different setup steps.
On this page
Choose a connection
| Use case | Connect with | Setup |
|---|---|---|
| Your application calls a model | API key + model API | Quickstart |
| Codex or Claude Code uses our models | API key + client provider settings | Coding agents |
| An agent needs inference and account tools | MCP browser sign-in or API key | MCP server |
| You want guided tools and commands | Client plugin + MCP sign-in | Plugins |
Base URLs
OpenAI SDK / Codex https://xavainf.com/v1 Claude Code https://xavainf.com Gemini CLI https://xavainf.com MCP with sign-in https://xavainf.com/mcp/account
Codex appends /responses to its base URL. Claude Code appends /v1/messages, so its base URL must not include /v1. A complete request URL belongs in an HTTP call, not in an SDK's base URL setting.
Keep environments together
Production uses xavainf.com; staging uses zinf.dev. Accounts, balances and keys belong to the environment where you created them. When switching environments, update both the hostname and the key.
Examples throughout these docs use the site you are viewing. Keep credentials in environment variables or your secret manager. Never include them in browser code or a shared configuration file.
Browse before authenticating
GET /v1/models is public. Use it to discover model ids, request endpoints and prices. Inference requests require an API key and available funds.