How can I use Codex with InferMatrixCopilot for PR reviews?

[hsliuustc0106]

Question

I would like to use Codex together with InferMatrixCopilot for read-only PR reviews.

The current MCP documentation shows that Codex can start and poll copilot workflows, but the example configuration forwards ANTHROPIC_API_KEY to infermatrix-copilot-mcp. Could you clarify the intended integration model?

1. When Codex invokes InferMatrixCopilot through MCP, is a separate ANTHROPIC_API_KEY (or another Anthropic-compatible endpoint key) always required?
2. Can InferMatrixCopilot reuse Codex's existing ChatGPT/Codex authentication, or is that authentication intentionally isolated from MCP servers?
3. Is Codex supported as the copilot's internal inference backend, for example through the Codex SDK, codex exec, or the OpenAI Responses API?
4. If Codex cannot currently be the internal backend, what is the recommended way to use Codex and still benefit from InferMatrixCopilot's playbooks, repository knowledge, run reports, checkpoints, and audit trace?
5. Could the documentation include a complete recommended Codex setup and example PR-review prompt?

My desired workflow is:

Codex -> InferMatrixCopilot -> read-only PR review -> Codex validates/summarizes the report

Ideally this would work with Codex signed in through a ChatGPT subscription, without copying cached Codex credentials or configuring an additional model-provider key. If that is not supported by design, documenting the credential boundary would also be helpful.

Relevant documentation:

• doc/MCP.md
• docs/codex/config.toml
• src/infermatrix_copilot/llm.py

[hsliuustc0106]

cc @tzhouam