Execute a prompt via Codex CLI. Codex is an AI coding agent that can generate, analyze, refactor, and explain code with full project context (reads AGENTS.md/CODEX.md automatically).

Capabilities: code generation and refactoring, code analysis and explanation, file reading and modification (when sandbox allows), git operations and terminal commands, multi-turn conversations via sessionId.

When to use a different tool:

• For analysis of text you already have (plans, docs, opinions), inline it directly in the prompt rather than passing file paths. The files parameter triggers full file I/O and increases timeout pressure.

Tips:

• Set workingDirectory to the target repo for project-aware responses.

• Use sandbox "read-only" (default) for analysis, "full-auto" for code changes.

• Break complex tasks into focused prompts rather than one large request.

• Resume multi-turn conversations with sessionId (returned in previous response metadata).

• Include relevant files via the files parameter for targeted context (text and images supported).

• Set reasoningEffort to control depth: "none" for trivial, "minimal" for lightweight, "low"/"medium" for routine, "high"/"xhigh" for deep analysis.

Web search via Codex CLI. Searches the web and synthesizes a comprehensive answer with source URLs. Use for current information, documentation lookups, API references, and research questions.

Tips:

• Ask specific, focused questions for best results.

• Results include source URLs for verification.

• Use maxResponseLength to control response verbosity.

• Default timeout is 2 minutes; increase for complex research queries that may require multiple web fetches.

Lightweight query for analysis or opinions on text you already have. No file reading, no repo exploration, no session state. Pass text in the context parameter. For code execution or file operations, use the codex tool instead.

Use cases: reviewing a plan, critiquing a draft, comparing approaches, answering questions about provided text, generating summaries.

Tips:

• Pass the text to analyze in the context parameter, not as file paths.

• Set reasoningEffort to control depth: "none" for trivial, "minimal" for lightweight, "low"/"medium" for routine, "high"/"xhigh" for thorough analysis.

• Use maxResponseLength to control verbosity.

Run Codex CLI's native diff-aware review via codex exec review --json. The bridge does not accept or bundle a review prompt; upstream Codex owns the reviewer instructions.

Use for MCP clients that cannot run shell commands but need native Codex review of uncommitted changes, a base branch diff, or a single commit.

Tips:

• Set workingDirectory to the target git repository.

• Choose mode "uncommitted", "base", or "commit".

• Use the codex tool with sandbox "read-only" for free-form review prompts.

Generate a JSON response conforming to a provided JSON Schema. Use for data extraction, classification, or any task needing machine-parseable output.

List active Codex conversation sessions. Returns session metadata for orchestration (no prompts or responses, just IDs and timing). Use to check available sessions before resuming with the codex tool's sessionId parameter.

Report what Codex's own app-server says about each MCP server it knows about: auth type, tool inventory, and whether the server initialized.

Unlike the other tools, this one deliberately does NOT suppress Codex's MCP servers — suppressing them would disable the thing being measured — so it is slower than a normal call and boots the servers named in ~/.codex/config.toml.

Two modes:

• Default (~6-9s): inventory only. Reports each server as "initialized" or "unknown". Creates no thread and writes no session record.

• diagnostics: true (~10-20s): starts an ephemeral thread to collect startup notifications, which are the only source of an explicit "failed" state and the error text naming the cause (expired OAuth grant, missing binary, remote refusal).

Reading the output:

• "unknown" is not a failure. It means the inventory carried no server info and no explicit verdict was available. Only diagnostics mode can report "failed".

• A "degraded" warning means the underlying call was slow enough that healthy servers have been observed reporting as uninitialized. Treat "unknown" as unproven when it appears.

• Servers marked "builtIn" are injected by Codex and are not in your config.toml. Servers marked "configuredButUnreported" are in your config.toml but absent from Codex's inventory.

Health check: verifies Codex CLI is installed and authenticated, reports versions and capabilities.