Codex Vision MCP

MCPReference

by darkamenosa

Indexed by Glama. No runnable package is published via Glama's API — open the repository for install instructions.

mcplocal-only

MCP configuration

stdio

Environment

  • CODEX_VISION_MODEL

    Responses model to use for vision analysis. Default: gpt-5.5

  • CODEX_VISION_BASE_URL

    Base URL for the Codex Responses backend. Default: https://chatgpt.com/backend-api/codex

  • CODEX_VISION_AUTH_JSON

    Explicit path to a JSON file containing OAuth credentials. If not set, the server attempts to discover auth from other sources.

  • CODEX_VISION_TIMEOUT_MS

    Timeout in milliseconds for API requests. Default: 300000

  • CODEX_VISION_IMAGE_DETAIL

    Image detail level: auto, low, high, or original. Default: auto

  • CODEX_VISION_MAX_IMAGE_MB

    Maximum image file size in MB. Default: 10

  • CODEX_VISION_MAX_VIDEO_MB

    Maximum video file size in MB. Default: 50

  • CODEX_VISION_MCP_LOG_PATH

    Optional path to a log file for the MCP server.

  • CODEX_VISION_VIDEO_FRAMES

    Number of frames to extract from video. Default: 4

Install in one click with Zurek

Browse here, install from the native macOS app — pick targets, fill env vars, and write configs safely.