About ask-codex
Ask LLM connects coding assistants and MCP clients to OpenAI Codex CLI for second opinions on code and plans. It is aimed at developers who want another model to review changes or challenge a plan from within their existing assistant session.
Key features
- 01Review code and staged changes
- 02Challenge plans and assumptions
- 03Use authenticated provider subscriptions
- 04Opt-in per-edit pair reviews
- 05Read-only provider controls for reviews
Use cases
- 01Check staged changes for bugs
- 02Ask Codex for a security review
- 03Compare approaches before implementation
MCP support
Exposes review and plan-challenge tools to MCP clients, connecting them to Codex CLI and other supported providers. Setup configures host registrations and workflows.
Configuration
Local server config is the same for every client. Copy it into your client’s MCP config file.
npx -y @ask-llm/codex-mcpGMCPT_TIMEOUT_MSTimeout for Codex CLI execution in milliseconds (default: 300000 = 5 minutes)GMCPT_LOG_LEVELLog verbosity: debug, info, warn, error (default: warn)Local servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.
Tools
4Saved tool list · updated 2w ago
| Parameter | Type | Required | Description |
|---|---|---|---|
| model | string | No | DO NOT set this parameter. The tool automatically uses gpt-6-astra and falls back to gpt-5.6-terra on quota errors. Only set this if the user explicitly requests a specific model. |
| prompt | string | Yes | The question, code review request, or analysis task to send to Codex CLI |
| sandbox | enum | No | Codex sandbox mode for this call. Defaults to 'read-only', which enforces the core review contract (Codex reads and proposes, the MCP client edits). Set 'workspace-write' ONLY as an explicit opt-out for flows that need Codex to write files itself, e.g. image generation. Review, second-opinion, and analysis flows must never set this.read-only | workspace-write |
| preferred | boolean | No | Opt into ASK_CODEX_PREFERRED_MODEL when it is configured to differ from the gpt-6-astra default. The built-in preferred value is also gpt-6-astra, so normal calls and review skills should leave this unset. |
| sessionId | string | No | Omit for an ephemeral one-off call. Pass an empty string ("") on the first call to create a persisted resumable thread; then pass the returned [Thread ID: ...] value on follow-up calls to continue with full prior context. |
| includeDirs | array | No | Additional directories Codex may access alongside the working directory on a fresh call (maps to codex `--add-dir`, repeatable). Rejected together with a resumed sessionId because `codex exec resume` has no --add-dir; directories are never silently dropped. Must be relative paths (e.g., 'packages/api'). Useful in monorepos where relevant context spans sibling packages. |
| reasoningEffort | enum | No | Codex reasoning effort for this call. Defaults to medium; /codex-review and /brainstorm use high for quality-first work.low | medium | high | xhigh | max | ultra |
Details
From the project READMEAsk LLM
Second opinions for coding agents.
Get a second opinion on your code or plan, inside your coding assistant, using the model subscriptions you already have.
For individual developers and small teams who already use a coding assistant and pay for another model subscription: ask a different vendor's model to review your changes or challenge your plan, without API keys, copy-paste, or leaving the session.
You: ask codex to review src/auth.ts for security issues
Codex: ⚠ verifyToken() compares tokens with === — not timing-safe (line 42)
⚠ the session cookie is missing a SameSite attribute
Claude: Good catches — applying both fixes to src/auth.ts.
Install
You need Node.js 24+ on Linux or macOS. Before your first review, both your coding assistant (for example Claude Code) and the provider you will consult (for example Codex) must be installed and signed in. Ask LLM configures the connection; it does not install providers or sign you in. If a provider CLI such as Codex has never been signed in on this machine, sign in first and rerun setup.
npm i -g @ask-llm/mcp && ask-llm setup
Setup asks you to confirm each detected host and workflow. Inspect the preview, confirm your assistant's registration and workflows, decline the others, then follow any printed manual steps. Restart your assistant or start a new session in your project.
Your first review
With both Claude Code and Codex installed and signed in, open Claude Code in your project and ask:
Ask Codex to review my staged changes for bugs and missing error handling. Check each finding against the source before proposing fixes.
Working in Codex with a Claude subscription? Ask Claude instead. Choose a provider from a different vendor than your current assistant. A successful review names the reviewer and model, cites relevant code, and reports findings or says it found none. Your assistant applies changes after you decide what to accept. If the request fails, run ask-llm doctor and follow the troubleshooting guide for your selected assistant and reviewer. Warnings about providers you do not use do not require installing them or adding API keys.
For plans, try: “Ask Claude to challenge this plan: what assumptions could fail, and what simpler approach would fit our constraints?”
After an on-demand review, pair review can add ongoing feedback. Pairing requires explicit opt-in; setup leaves it off.
Supported hosts
A host is the assistant you work in; a provider is the model you consult.
| Host | Experience |
|---|---|
| Claude Code | Reviews, plan critique, comparisons; isolated reviewers and opt-in per-edit pairing |
| Codex, Cursor, Grok Build, Gemini CLI | Tools and portable workflows; pairing on demand |
| Antigravity | Tools; setup prints a manual step for workflows |
| Pi 0.83.0+ | Native tools and portable workflows; opt-in per-edit pairing |
| Claude Desktop | Tools only |
| OpenCode | Tools and portable workflows; setup verified with fixtures only |
See host compatibility for the complete matrix and manual setup limits, including existing Codex registrations.
Supported providers
| Provider | Access |
|---|---|
| Codex | Your authenticated Codex CLI/account |
| Claude | Your authenticated Claude Code account; consult it from a host other than Claude Code |
| Grok | Authenticated Grok Build, an exact Cursor model, or the explicit metered xAI API option |
| Antigravity | Your authenticated agy subscription; execution is off by default, and opt-in has no read-only guarantee |
| Ollama | Local models on your machine |
| Gemini | An eligible Gemini Code Assist Standard/Enterprise seat |
Cursor Agent can also consult exact models from your Cursor catalog. It is a connection route, with usage governed by your Cursor plan.
Trust and cost
Review runs use read-only provider controls; the assistant you work in remains the editor. Antigravity refuses execution until explicitly enabled and cannot guarantee read-only access after opt-in. Review only material you are allowed to share with the selected provider.
Calls consume your existing provider plan's quota or follow that account's billing. Grok's xAI API is a separate, metered option requiring an API key; Ask LLM never enables billing or overage for you. Ollama supports local inference. Ask LLM sends no telemetry; provider and host data policies still apply.
Agreement between models is not proof of correctness. Validate findings against the code, tests, and your requirements. See data flow and costs and fallbacks.
Migrating from @ask-llm/*
Follow the migration guide before retiring earlier registrations.
Upgrade or remove · Troubleshooting · Documentation
Contributions: CONTRIBUTING.md. Licensed under MIT.
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.