OpenAI Codex
OpenAI's coding agent across a CLI, IDE extension, the ChatGPT app and Codex cloud, for writing, reviewing, debugging and automating software work.
Coding agent · OpenAI
OpenAI's coding agent across a CLI, IDE extension, the ChatGPT app and Codex cloud, for writing, reviewing, debugging and automating software work.
Harnesses and platforms · Individual models. Model availability, orchestration and connector access depend on the specific product and surface; one does not imply the others.
The five-point review
Judgment, not a score. Documentation checked 2026-10-18.
- Value
- OpenAI's coding agent across a CLI, IDE extension, the ChatGPT app and Codex cloud, for writing, reviewing, debugging and automating software work.
- Differentiation
- Delegating scoped coding tasks and reviewing diffs; Long or parallel background tasks in Codex cloud; Teams already standardized on ChatGPT; Sharing one MCP configuration across the CLI, IDE and ChatGPT desktop app. These are documented fits, not exclusive advantages.
- Replaceability and switching
- Compare alternatives in the harness directory. Local files, cloud environments, connectors and saved conversation state may need separate migration; no drop-in replacement is established here.
- Ease and technical aptitude
- Cloud tasks do not automatically inherit local files, services, personal skills or MCP configuration. Configure and test the cloud environment’s repositories, network destinations and service access separately.
- Interoperability and portability
- The desktop app, CLI and IDE share MCP configuration for the same Codex host. ChatGPT web uses remote MCP tools through plugins; it does not read local config.toml. Connector access does not establish export or full workspace portability. See the surface-specific limits below.
Connections: MCP connections · APIs · Portability and vendor lock-in.
Related individual models: GPT-6 Astra · GPT-6.1 Sol · GPT-6 Luna. These links explain model choices; availability in this app or harness must be checked separately.
What it is
Codex is OpenAI's AI coding agent. It runs locally through the Codex CLI and IDE extension, inside the ChatGPT desktop app, and in Codex Cloud, where each task runs in its own isolated workspace so work can continue while your computer sleeps. It reads a codebase, edits files and runs commands.
- Codex CLI: Terminal agent source ↗
- IDE extension: Editor-based edits source ↗
- Codex Cloud: Isolated workspaces for background tasks source ↗
What it's best at
- Delegating scoped coding tasks and reviewing diffs
- Long or parallel background tasks in Codex cloud
- Teams already standardized on ChatGPT
- Sharing one MCP configuration across the CLI, IDE and ChatGPT desktop app
How it works
- Locally, the CLI and IDE extension read your repo, edit files and run commands under configured permissions and approval policies.
- Codex Cloud runs each task in an isolated workspace built from a cloud environment you prepare; internet access is a setting you control.
- Configure STDIO or Streamable HTTP MCP servers in ~/.codex/config.toml or a project .codex/config.toml. The desktop app, CLI and IDE share configuration for the same Codex host; cloud setup is separate.
- Browser is available in the ChatGPT desktop app, not in the Codex CLI or IDE extension.
Agent, connector and MCP support
Ideal use cases
- Clear a backlog of small, well-specified issues in parallel
- Write tests and docs for existing code
- Pull PostHog or GitHub context into a coding task through MCP
- Prepare a deploy to Railway with the Railway Codex plugin
Limits
- Cloud tasks do not automatically inherit local files, services, personal skills or MCP configuration. Configure and test the cloud environment’s repositories, network destinations and service access separately.
- ChatGPT Work and Codex share usage limits; check what your ChatGPT plan includes.
- Every change still needs review and tests.
- The built-in browser cannot automate file uploads; attaching files to a chat or creating files in a workspace is a separate capability.
How it pairs with your tools
- GitHub: GitHub MCP server install guide for OpenAI Codex
- PostHog: Named as a supported MCP client
- Railway: Railway Codex plugin
- FactSet: Codex CLI listed as an MCP client
Also named as a supported client in the official docs of: 1Password, Apify, Apollo.io, Brex, CapCut, Chatbase, Clay, Cloudflare Workers, Consensus, Descript, Figma, Hostinger, Langfuse, lemlist, Mercury, Notion, OpenRouter, OpusClip, PandaDoc, Readwise, Remotion, Surfer, Tidio, Twilio, Vercel, Wave (AI note-taker), Webflow, Wispr Flow, Wolfram.
Tool-page matrices include OpenAI Codex and distinguish documented routes, inferred combinations and unverified capabilities. Check the cited sources and requirements before connecting.
Workflows you can run here
Confirmed means the workflow's connection is documented for OpenAI Codex, including a native recipe explicitly scoped to this platform. Inferred means an API step relies on OpenAI Codex running code, with no document naming the pairing. These labels do not mean IonBin has tested the workflow.
- Turn Stripe events into team tasks
- Route requests from a form to the right board
- Protect the machines and accounts your agents use
- Run an AI support agent that can take actions
- Use AI for first responses in live chat
- Update your privacy policy for AI features
- Organize tax documents with Claude, then file in FreeTaxUSA
- Switch tax software using last year’s return
- Sanity-check investment tax forms before you upload them
- Ask questions in ChatGPT, then file in TurboTax
- Compare tax software tiers with cited sources
- Research company formation options with citations
- Compare contractor vs employer-of-record costs
- Decide between a PEO and payroll-only
- Write benefits enrolment messages for HR approval
- Analyze survey results
- Build a checked contract register
- Capture a process as a reviewed SOP
- Draft policies from approved practice
- Prepare a year-end accountant pack
- Redact documents before AI review
- Score AI drafts with two reviewers
- Route website leads straight to the right calendar
- Deliver a lead magnet automatically
- Keep sensitive documents out of AI chats
- Launch a website support agent with Botpress
- Qualify leads with a bot, then book the call
- Triage support tickets with an agent team
- Ask several AI models and reconcile the answers
- Package your best prompts as reusable agent skills
- Write the script and captions, edit in CapCut
- Turn trend research into a CapCut edit
- Rewrite your custom instructions from your own chat history
- Ask a mutual connection for an intro, with a scheduled follow-up check
- Write a policy brief grounded only in your documents
- Turn long videos and papers into cited notes
- Build an onboarding notebook new hires can question
- Shortlist used-car listings against book value
- Monitor competitor pricing pages with Gumloop
- Build a shortlist of small businesses to approach from public records
- Talk through a brief, get a clean document
- Clear email by voice with an AI editor
- Make your offer page readable by AI agents
- Audit a role before you hire for it
- Get a weekly repo report from a scheduled agent
- Switch AI assistants and keep your context
- Check your agent setups for the lethal trifecta
- Build an evaluation set from support tickets
149 more with at least one step To verify
- Give every AI tool your brand context
- Turn a script into an avatar video from chat
- Add an AI voiceover to a video
- Publish audio editions of your articles
- Dub and localize product videos
- Cut social clips from a long video
- Draft, optimize and publish an SEO article
- Audit how AI answers mention your brand
- Turn keyword research into a content plan and tasks
- Write a monthly SEO and product report
- Fix site-audit issues as pull requests
- Create directory or landing pages from a Notion database
- Build a landing page with a coding agent in Framer
- Add a writing-quality gate before publishing
- Build a presentation from research
- Make a tailored one-pager for each prospect
- Turn a blog post into an email campaign
- Ask product analytics questions from your editor
- Run an A/B test with an AI pair
- Triage new leads and draft follow-ups
- Log meeting notes and next steps in your CRM
- Write a weekly pipeline forecast
- Get next-step coaching on open deals
- Build a qualified prospect list
- Run an outbound sequence with approval
- Triage outreach replies automatically
- Manage cold-email campaigns from chat
- Prepare a pre-call account brief
- Enrich new leads automatically
- Draft proposals and chase signatures
- Give your AI agent live web data
- Monitor competitor and vendor pages for changes
- Check search rankings and sample AI answers
- Ask questions across your meeting history
- Let your AI agent trigger your automations
- Follow up on failed payments
- Ask revenue questions in plain language
- Ask bookkeeping questions in QuickBooks
- Check Stripe payouts against QuickBooks
- Turn Fireflies action items into ClickUp or monday.com tasks
- Write a weekly project status update
- Hand coding tasks to an agent from your board
- Ship a small AI tool to production
- Self-host n8n or your own MCP server
- Use secrets with an agent while limiting exposure
- Audit SaaS and AI tool spend
- Turn support conversations into weekly insights
- Add an AI agent to your website
- Draft support replies for agent approval
- Answer calls with an AI voice agent
- Ask HR and payroll questions in plain language
- Onboard a new hire or contractor
- Invoice when work is done
- Write a weekly cash and spend note
- Keep a compliance and filings calendar
- Transcribe and summarize calls automatically
- Check estimated-tax arithmetic with Wolfram in Claude
- Hand clean books to your tax filing
- Run month-end reconciliation with Xero in Claude
- Draft a Xero invoice when a project is done
- Turn Xero data into a cash forecast and investor update
- Check Stripe payouts against Xero
- Review card spend for policy issues
- Take an agreement from template to signature
- Send the agreement when a deal is won
- Turn an intake form into a signed consent form
- Run a founder launch checklist in ClickUp
- Review unsigned agreements and draft Docusign reminders
- Build a brand-name shortlist and check trademarks
- Plan an international hire with Remote
- Set up payroll with a checklist
- Analyze an export by chatting with it
- Double-check key figures with Wolfram
- Power a calculator with the Wolfram LLM API
- Pull data lookups into research notes
- Let Claude schedule focus time for your tasks
- Prepare a recording library for retrieval
- Turn sprint issues into booked time
- Book and reschedule meetings from chat
- Turn a recorded call into a follow-up
- Find themes across customer interviews
- Build an intake form from chat
- Send form submissions to your CRM and score them
- Run a survey into Airtable or Notion
- Launch a site from chat with Hostinger
- Take your site multilingual
- Run n8n on a VPS managed from chat
- Publish Claude drafts to WordPress on Kinsta
- Register a domain and point it at Cloudflare Pages
- Fix weak passwords from a health report
- Write articles that cite real research
- Cross-check web answers against research
- Turn your reading highlights into an outline
- Build an evidence table from a literature review
- Summarize last week’s calls into CRM notes
- Turn Otter meeting action items into ClickUp tasks
- Add accurate captions to marketing videos
- Draft an on-brand campaign with Jasper
- Write product descriptions in bulk for Webflow
- Capture research into a knowledge base agents can read
- Check factual claims before you publish
- Make press releases easy for AI answers to quote
- Write a weekly ad report from connected account data
- Run social posts through an approval queue
- Get a daily brief from your tools
- Compute finance numbers with code, not chat
- Give agents read-only access to your notes
- Automate a web task that has no API
- Check deploys and logs from your coding agent
- Render many video variants from data
- Rough-cut in Descript, polish in CapCut
- Turn a podcast into shorts from your coding agent
- Turn webinar recordings into clips for review
- Rewrite clip titles and captions per platform
- Schedule clips from chat, with a human check
- Edit a podcast with Descript from your AI assistant
- Answer the phone with an intake agent, then warm-transfer to a person
- Edit the long version in Descript, clip it with OpusClip
- Turn sales calls into an account brief
- Research with Grok Bot, record and edit in Descript
- Find weekly themes across customer calls
- Build a quiz funnel and A/B test each question with PostHog
- Send agent-drafted letters through a mail API, with approval
- Control which provider runs your AI model
- Give your AI a small database of your own context on Railway
- Have Claude run a Gumloop lead-research agent
- Follow up with quiet leads by sending one useful thing, no ask
- Report ad results by profit, not revenue
- Set ad targets from break-even ROAS
- Re-run your evaluation set when a model changes
- Turn a PDF folder into checked, source-linked Markdown
- Check last week's Google Ads results against site analytics
- Turn recurring Intercom support themes into Linear issues
- Turn a Confluence meeting page into reviewed Jira work items
- Answer weekly KPI questions from Metabase with an AI client
- Review next week's Calendly bookings and availability with an AI client
- Prepare for new Calendly bookings with HubSpot context
- Review last week's Klaviyo campaigns and flows with an AI client
- Draft a Klaviyo email campaign with an AI client, then review it before sending
- Get a daily Square sales summary from an AI client
- Find help-article gaps from Help Scout conversations
- Create Freshdesk tickets from a reviewed list with an AI client
- Run a weekly SEO check with Ahrefs MCP and Search Console
- Turn a Figma frame into code with the Figma MCP server, then review it
- Ask product questions of your Supabase database in read-only mode
- Let an AI client run approved follow-up actions through Zapier MCP
- Summarize sensitive documents locally with Ollama
- Check LlamaParse output on a sample before parsing a whole folder
- Prototype an internal tool in Lovable through MCP, and check access before publishing
In the wiki
Sources
- Codex developer workflows · OpenAI · official · accessed 2026-10-18
- Model Context Protocol (ChatGPT Learn) · OpenAI · official · accessed 2026-10-18
- Browser · OpenAI · official · accessed 2026-10-18
- Pricing · OpenAI · official · accessed 2026-10-18
- Codex Cloud environments · OpenAI · official · accessed 2026-10-18
To verify Codex usage limits for each ChatGPT plan (checked 2026-10-17)