Agents and developer clients
Choose the client protocol before connecting Codex, Claude Code, or Cursor to XiuRouter
Codex, Claude Code, and Cursor use different setup paths and API formats. Choose the client below, then follow its guide and verify one real task.
Choose by client, not model brand
The client determines the API format and setup. The selected model must support that format and be allowed by the API key.
- Use for
- Local Codex work in the CLI, IDE extension, or desktop app.
- Choose this path
- Choose Responses, configure a XiuRouter provider for new local tasks, and verify it with one small read-only task.
- Limits
- This changes local Codex requests, not cloud ChatGPT chats. Provider-hosted retrieval, stored conversations, and background mode are outside the current scope.
- Use for
- Claude Code CLI, the VS Code extension, and the desktop Code interface.
- Choose this path
- Choose Anthropic Messages and use the gateway root without duplicating /v1. Confirm success only after a local task completes.
- Limits
- A local gateway does not add Claude cloud environments, SSH, Remote Control, or Anthropic-hosted server tools.
- Use for
- Cursor custom models when the OpenAI Base URL can be changed.
- Choose this path
- Choose Chat Completions and verify Ask plus the request record first. Test model-specific tool calling separately before relying on Agent edits.
- Limits
- The Base URL override is global. Cursor-specific models and Tab completion stay on Cursor’s service, and compatible routing alone cannot guarantee Agent edits.
Verify the client workflow, not only a text reply
Test the smallest real task you depend on, then check its Usage record.
- 1
Confirm the client can reach the expected protocol path with the selected model.
- 2
For coding agents, test one bounded tool or file task instead of only a text reply.
- 3
Check Usage for the model, tier, status, tokens, and recorded cost before relying on the setup.