MCP server that bridges ChatGPT with Codex CLI, enabling code analysis, planning, and application with session management and project context support
Codex Bridge defines 16 tools with consistent naming (verb_noun pattern: codex_*) and complete input schemas. All tools have descriptions in Chinese, though many are brief (20-80 chars, below the 50-200 char LLM-optimized baseline). Parameter descriptions are present but terse. Output schemas are not documented in the visible code. Tool annotations (readOnlyHint, destructiveHint, idempotentHint) are correctly applied to codex_status. Error handling is minimal, no recovery guidance or actionable error messages visible. The server lacks composition patterns for multi-step workflows (e.g., prepare_apply → apply requires manual token threading). No pagination support visible for list_sessions or list_projects despite potentially large result sets.
向 Codex 提交分析任务,获取代码分析结果。
应用代码修改到项目,需要提供有效的确认令牌。
取消正在执行的异步任务。
继续已有的 Codex Desktop 会话,执行新的回合。
在 Codex Desktop 中创建新的持久会话。
将 ChatGPT 当前会话上下文交接给 Codex,用于首轮请求。
获取已完成任务的结果。
Descriptions are too brief (20-80 chars vs. 50-200 char LLM-optimized baseline). Most lack context on WHEN to use the tool or what it returns. E.g., 'codex_analyze' is '向 Codex 提交分析任务,获取代码分析结果' (40 chars), does not explain what 'analysis' means or when to call it instead of codex_plan.
Output schemas are not documented. The visible code shows input schemas but no return type documentation. LLMs cannot plan downstream calls or extract required fields (e.g., does codex_list_projects return project_id, name, root_paths?). The _project_payloads helper suggests a structure, but it is not exposed in tool definitions.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 54 | <=2025-11-25 | v2 |
查询异步任务的当前状态。
列出所有已授权项目,包括手动配置的根目录和 Codex 项目登记中的项目。
列出 Codex Desktop 侧边栏中的可见会话。
向 Codex 提交规划任务,获取代码修改方案。
预检查并准备代码应用,返回确认令牌供后续 codex_apply 使用。
为指定项目收集会话上下文,包括历史消息和相关信息,用于 ChatGPT 理解项目背景。
为受限会话生成临时访问令牌,用户需确认后才能读取。
读取指定 Codex Desktop 会话的消息内容。
检查本机 Codex CLI;会话快照仅在调用方显式请求时返回。
No pagination support visible for list_sessions (accepts cursor and limit but no documented total_count or next_cursor in response). list_projects has no pagination at all. Large result sets risk context window exhaustion.
Error handling is absent from visible code. No recovery guidance, error classification, or actionable messages. E.g., if codex_apply fails due to invalid confirmation_token, the LLM has no guidance on whether to retry, call codex_prepare_apply again, or ask the user.
Tool composition requires manual token threading (codex_prepare_apply → codex_apply). No confirmation pattern visible in code; agents must infer the workflow. Consider wrapping into a single idempotent tool or documenting the multi-step flow explicitly.