TL;DR: You can run Claude Code and Codex together without a special “bridge” product: install both CLIs, sign into Claude Pro (or higher) and ChatGPT Plus (or higher), and keep one shared instruction file. Codex reads AGENTS.md; Claude Code can load the same AGENTS.md when you have no CLAUDE.md, or both files if you set Project instructions to claude-md-and-agents-md. A practical split is Claude Code for long agent loops and Codex for a second-model review (or the reverse). Two entry plans are about $20 + $20 a month (UK App Store: Claude Pro £20, ChatGPT Plus £19.99).
Last updated: November 6, 2026. By Veer. Setup steps and prices checked on November 6, 2026 against Claude Code docs, Claude pricing, Anthropic’s Claude Code + Pro/Max/Ultra help, Codex CLI docs, Codex pricing, and AGENTS.md guidance. UK App Store IAP prices from Apple’s ChatGPT and Claude UK listings. This is a researched how-to built from public docs and pricing pages — TrustLoud has not run a timed dual-agent trial on a private codebase.
Developers searching how to use Claude Code and Codex together usually want a dual-CLI workflow, not a single merged product. Claude Code (Anthropic) and Codex (OpenAI) are separate terminal coding agents. Neither vendor ships an official “run both as one agent” product. What is documented is parallel install, shared project instructions, MCP on both sides, and independent usage pools — enough for a clean dual-agent workflow if you treat them as two specialists on the same repo.
For a pure price-and-limits shootout, see TrustLoud’s Claude Code vs Codex comparison. This guide is the workflow playbook: when to run both, how to set them up, how to share instructions, and what the dual-subscription bill looks like.
How to use Claude Code and Codex together: when it makes sense
Use both when you want a second model on the same codebase without switching subscriptions mid-sprint — for example one agent implements, the other reviews, or you keep working when one plan hits its five-hour window. Autocomplete demand for “claude code and codex working together” tracks that pattern: complementary roles, not a single merged agent.
Good fits:
- You already pay for Claude Pro (or Max/Ultra) and ChatGPT Plus (or Pro) for chat, so the coding agents are already included.
- You want cross-vendor review: one agent writes the diff, the other runs
/reviewor a PR pass. - One vendor’s five-hour or weekly limit keeps biting; the other pool is still open.
- Your team already standardized on AGENTS.md for other tools, and you want Claude Code to read the same file.
Skip the dual setup when one agent covers your repo size and hours, or when paying ~$40/month for two entry plans is wasteful. In that case pick from the Claude Code vs Codex table and stay on one side.
Set up Claude Code (Anthropic)
Claude Code is Anthropic’s agentic coding tool for the terminal, IDE, desktop app and web; Pro and higher include it, Free does not. Official install (macOS, Linux, WSL) from the Claude Code overview:
curl -fsSL https://claude.ai/install.sh | bash
cd /path/to/your/project
claude
Windows PowerShell: irm https://claude.ai/install.ps1 | iex. Homebrew (brew install --cask claude-code) and WinGet (winget install Anthropic.ClaudeCode) are also documented; native installs auto-update, Homebrew/WinGet do not.
On first launch, sign in with a Claude Pro, Max, Ultra, Team or Enterprise account, or a Claude Console / cloud-provider path (quickstart). Anthropic’s help center confirms Pro/Max/Ultra unlock Claude Code in the terminal and supported IDEs (VS Code, Cursor, JetBrains), and that IDE usage shares the same limits as Claude chat (help article).
Project memory defaults to CLAUDE.md (run /init to draft one). Claude Code can also read a repo AGENTS.md — see the shared-instructions section below (memory docs).
Set up Codex (OpenAI)
Codex is OpenAI’s local coding agent (CLI, IDE extension, web, iOS) that inspects the repo, edits files, runs commands and can review diffs. Install from the Codex CLI docs:
curl -fsSL https://chatgpt.com/codex/install.sh | sh
cd /path/to/your/project
codex
Windows: irm https://chatgpt.com/codex/install.ps1 | iex. npm (npm install -g @openai/codex) and Homebrew cask installs are listed on the openai/codex repo.
Sign in with ChatGPT (Plus, Pro, Business, Edu or Enterprise for full Codex surfaces) or an API key for usage-based CLI/IDE work without cloud features (authentication, pricing). Inside the TUI, /init creates an AGENTS.md; /permissions and sandbox settings control what Codex may edit or network to (approvals & security). Defaults commonly keep network off and writes scoped to the workspace unless you widen them.
Share one instruction file both agents can read
The documented bridge is a shared markdown instructions file — not a live sync between the two products. Codex always builds an instruction chain from AGENTS.md / AGENTS.override.md under ~/.codex and the project tree (AGENTS.md guide). Claude Code’s default is CLAUDE.md, but Anthropic’s memory docs say Claude Code can load AGENTS.md in place of CLAUDE.md when no CLAUDE.md / CLAUDE.local.md sits on the project path (Claude Code v2.1.277+), or load both if you set Project instructions to claude-md-and-agents-md in /config (memory / AGENTS.md).
Practical patterns that match the docs:
- AGENTS.md only — put build commands, lint rules and “never do X” notes in
./AGENTS.md. Codex reads it; Claude Code reads it when no CLAUDE.md is present. - CLAUDE.md that imports AGENTS.md — keep Claude-specific notes under an
@AGENTS.mdimport so both tools stay aligned even when a CLAUDE.md exists. - Both files on purpose — set Claude’s Project instructions to
claude-md-and-agents-md, keep shared rules in AGENTS.md and Claude-only prefs in CLAUDE.md.
Do not assume AGENTS.override.md, Codex config.toml, or Claude .mcp.json are interchangeable. Those stay product-specific. Confirm loads with Claude’s /context or /memory, and with Codex by asking it to summarize active instruction sources.
A dual-agent workflow that stays sane
Run the agents in separate terminals (or IDE sidebars), give them different jobs, and commit between handoffs so neither clobbers the other’s half-finished edit. Official docs support MCP, review modes and instruction files on each side; they do not describe a vendor-blessed multi-agent bus between Claude Code and Codex.
A workflow that sticks to documented capabilities:
| Step | Claude Code | Codex |
|---|---|---|
| 1. Orient | claude → “what does this project do?” |
codex → same question, or /init if AGENTS.md is missing |
| 2. Plan | Ask for a short plan / impact list | Optional second opinion on the plan |
| 3. Implement | Apply the change; use Shift+Tab permission modes as needed |
Stay idle, or work a different path/branch |
| 4. Review | Optional local review | /review against uncommitted changes or a base branch (CLI) |
| 5. Ship | Commit / open PR | Tag cloud/GitHub review features if your ChatGPT plan includes them |
Role swaps work too: Codex implements, Claude Code reviews. Parallel work on separate git worktrees or branches avoids write conflicts. Both products document MCP for external tools (Claude MCP, Codex codex mcp in the CLI docs) — useful for issue trackers — but wiring the same MCP server into both is still two separate configs, not a shared runtime.
For model-tier choices inside Anthropic’s lineup while you are in Claude Code, see Sonnet 5.5 vs Opus 5.5. For “is Max worth it on the Claude side only,” see Claude Max 5x vs 20x.
What the dual setup costs (USD + UK App Store)
Expect about $40/month at the entry tier if you buy Claude Pro and ChatGPT Plus on the web; UK App Store IAPs are close in pounds and can run higher on Max/Ultra/Pro tiers. Claude Code needs Pro or above (claude.com/pricing); useful Codex CLI/IDE/web access is on Plus and above (Codex pricing).
| Piece | US web (official) | UK App Store IAP (6 Nov 2026) | What you get for coding |
|---|---|---|---|
| Claude Pro | $20/mo, or $17/mo billed annually ($200 upfront) | £20.00 / mo; annual £199.99 | Claude Code terminal + IDE; shared limits with Claude chat |
| Claude Max | $100/mo | £119.99 / mo | 5× Pro usage (5-hour + weekly); Claude Code |
| Claude Ultra | $200/mo | £249.99 / mo | 10× Pro weekly; no 5-hour limit on verified accounts |
| ChatGPT Plus | $20/mo | £19.99 / mo | Codex web, CLI, IDE, iOS |
| ChatGPT Pro | From $100 / $200 / $500 | £88.90 / £200.00 / £444.99 | Higher Codex allowance; Pro has no five-hour limit per OpenAI’s Codex pricing notes |
| Dual entry stack | ~$40/mo | ~£40/mo | Both agents, separate usage pools |
Sources: claude.com/pricing, developers.openai.com/codex/pricing, UK App Store listings for Claude and ChatGPT. Web checkout currency can differ from App Store IAPs; taxes may apply.
API-key mode exists on both sides (Claude Console credits; OpenAI API pricing) if you prefer metered spend over a second subscription — but subscription sign-in is what the docs recommend for the full local+cloud feature set. Broader “which chat plan is worth it” context: best AI subscription for personal use.
Limits and gotchas when you run both
Each product meters its own pool; running both does not merge or transfer allowance. Anthropic: Pro/Max use a five-hour session limit plus a weekly limit shared across Claude chat and Claude Code; verified Ultra drops the five-hour cap (help). OpenAI: Plus publishes estimated local messages per five-hour window (for example 15–160 for GPT-6.1 Sol on Plus); Pro plans currently have no five-hour limit; local and cloud Codex share the allowance and weekly limits may apply (Codex pricing).
| Gotcha | Why it bites | Mitigation from the docs |
|---|---|---|
| Two CLIs editing one dirty tree | Overlapping writes / confused diffs | Separate branches or worktrees; commit between handoffs |
| CLAUDE.md hides AGENTS.md | Claude Code default skips AGENTS.md if CLAUDE.md exists | Import @AGENTS.md, or set claude-md-and-agents-md |
| Hitting a limit mid-turn | Session pauses while the other vendor is still fine | Switch agents; or buy credits / upgrade that side only |
| MCP configured once | Each tool has its own MCP config | Add servers in both Claude Code and Codex if both need them |
| Sandbox vs permission modes | Codex defaults are tighter on network; Claude uses session permission modes | Read /permissions (Codex) and Shift+Tab modes (Claude) before long unattended runs |
| “We connected them with one MCP” claims | No official Claude↔Codex bridge in either vendor’s docs | Treat community hacks as unverified; stick to shared files + git |
Editor tip: both ship VS Code / Cursor-friendly extensions (Claude VS Code, Codex IDE). Running both panels at once is fine; still keep one writer at a time on a given file set. Pricing for Cursor vs Copilot vs Claude Code is covered separately in Cursor vs GitHub Copilot vs Claude Code pricing.
When to use only Claude Code or only Codex
Stay on one agent if your hours fit one plan’s limits and you do not need a second-model review. Pick Claude Code when you already live in Claude for writing/research and want one Anthropic bill that includes the coding CLI. Pick Codex when ChatGPT is already your daily driver and you want OpenAI’s review presets, cloud tasks and ChatGPT-plan bundling. Upgrade only the side that hits limits — Max/Ultra on Anthropic, or ChatGPT Pro tiers on OpenAI — before you assume you need both forever.
FAQ
Can you use Claude Code and Codex together on the same repo?
Yes. Install both CLIs, open the same project in two terminals, and share instructions via AGENTS.md (and optional CLAUDE.md import). There is no official single-process integration; coordination is via git and shared markdown.
Do I need both Claude Pro and ChatGPT Plus?
For the documented subscription path, Claude Code needs Claude Pro or higher, and full Codex surfaces are aimed at ChatGPT Plus or higher. You can also use API keys on either side, with different billing and fewer cloud features.
Will one AGENTS.md work for both tools?
Codex is built around AGENTS.md. Claude Code can read AGENTS.md when no CLAUDE.md is present (v2.1.277+), or read both when Project instructions is set to claude-md-and-agents-md, or via an @AGENTS.md import from CLAUDE.md.
How much does running both cost per month?
About $40 on US web entry plans (Claude Pro $20 + ChatGPT Plus $20), or roughly £40 on UK App Store entry IAPs (Claude Pro £20 + ChatGPT Plus £19.99), before tax nuances and before any Max/Pro upgrades.
Can Claude Code call Codex (or vice versa) through MCP?
Both support MCP for external tools, but neither vendor’s primary docs describe an official MCP that turns one coding agent into a tool inside the other. Treat that as unsupported unless you build and maintain it yourself.
What if I already compared Claude Code vs Codex?
Use the pricing and limits comparison to choose a primary agent, then return here for the dual-agent setup steps and shared AGENTS.md patterns.
Sources
- Claude Code overview and quickstart — install, surfaces, login options.
- Claude Code memory / AGENTS.md and .claude directory — CLAUDE.md, AGENTS.md, settings.
- Claude pricing and Use Claude Code with Pro, Max, or Ultra — plan access and shared limits.
- Codex CLI, AGENTS.md guide, Codex pricing, authentication.
- openai/codex on GitHub — alternate install channels.
- UK App Store: ChatGPT, Claude — IAP prices checked 6 November 2026.
