OpenAI's Codex and Anthropic's Claude Code are agents that read files, call connected tools and follow written instructions. For sales work they share the parts that matter: MCP connections, reusable skills and approval controls. The practical differences are which plan includes them, where they run and how they ask before acting.
Both started as coding agents, and both are now used for research and writing work. This comparison sticks to what each vendor documents. It does not rank the underlying models, which change every few months.
We make Buena, which connects to both, so Buena is the setup example. Facts come from OpenAI's and Anthropic's documentation, checked on September 28, 2026.
What each one is
Codex is OpenAI's agent. It runs in the ChatGPT desktop app, in ChatGPT on the web, as a command-line tool, as an IDE extension and in Codex cloud. Claude Code is Anthropic's agent. It runs in the terminal, in IDE extensions, in a desktop app and on the web.
For a salesperson, both do the same kind of work: research an account from sources, check a list against your customer profile, draft a first message and prepare a campaign in the tools you connect.
Side by side
| Area | Codex | Claude Code |
|---|---|---|
| Where it runs | ChatGPT desktop app, ChatGPT on the web, CLI, IDE extension and Codex cloud | Terminal, IDE extensions, desktop app and the web |
| Plans | Included in ChatGPT Free, Go, Plus, Pro, Business, Edu and Enterprise, with more limited access on Free and Go; an API key works in the CLI and IDE extension, without cloud features | Pro, Max, Team, Enterprise or a Console account; not the free Claude plan |
| Adding an MCP server | codex mcp add, an entry in ~/.codex/config.toml, or Settings and then MCP servers in the desktop app; codex mcp login to sign in | claude mcp add --transport http, then /mcp to sign in |
| Standing instructions | AGENTS.md | CLAUDE.md, or AGENTS.md when a folder has no CLAUDE.md |
| Reusable skills | SKILL.md folders in .agents/skills or ~/.agents/skills | SKILL.md folders in .claude/skills or ~/.claude/skills |
| Approvals | Works in a sandbox by default and asks before editing outside the workspace or using the network; read-only and full-access modes are available | Manual mode asks before most edits, commands and network access; from version 2.1.283, terminal and VS Code sessions start in auto mode, where a separate model reviews actions |
What actually differs for sales work
- The plan you already have. Codex comes with ChatGPT plans, including business plans. Claude Code needs a paid Claude plan or a Console account. If your company already pays for one, start there.
- Where you'll use it. Both have a desktop app that needs no terminal. Codex also runs in ChatGPT on the web and in Codex cloud, which can start work from Slack or the web; Claude Code runs on the web at claude.ai/code.
- How it asks before acting. Codex's default sandbox asks before network access or edits outside the workspace. From version 2.1.283, Claude Code's terminal and VS Code sessions start in auto mode, where a second model reviews actions instead of asking you. Either way, make sure any step that sends comes back to you.
The shared parts matter more. Both connect to the same MCP servers, both follow the open Agent Skills standard for SKILL.md folders, and both read standing instructions from a Markdown file. A team can use both without maintaining two workflows.
One setup that works in both
Keep standing instructions in AGENTS.md: Codex reads it, and Claude Code reads it when a folder has no CLAUDE.md. Keep each skill in its own folder with a SKILL.md file, and copy it to .agents/skills for Codex and .claude/skills for Claude Code.
Then connect the same MCP servers to both. For Buena, these are the commands shown in your workspace under Settings, then Integrations, then AI Tools:
codex mcp add buena --url https://mcp.buena.ai/mcp
codex mcp login buenaclaude mcp add --transport http --scope user buena https://mcp.buena.ai/mcpAfter you sign in, either one can research prospects, reveal verified work emails and prepare draft campaigns in Buena. Neither connection activates a campaign or sends a message on its own: sending needs an explicit approval, and your agent's approval settings decide whether it asks you first. Keep that step asking you, or approve campaigns in Buena.
Which to choose
- Your company uses ChatGPT Business or Enterprise: start with Codex, which is already included.
- You pay for Claude Pro, Max, Team or Enterprise: start with Claude Code.
- You want work to run in the cloud and start from Slack or the web: compare Codex cloud with Claude Code on the web.
- You're unsure: set up both with the same AGENTS.md and skills, run the same account brief in each, and keep the one whose drafts you edit least.
For a step-by-step setup in Anthropic's tool, read Claude Code for sales. For the systems worth connecting to either, read MCP servers for sales.
Sources
- OpenAI: Codex, Codex pricing and Codex MCP
- OpenAI: AGENTS.md, Codex skills and sandboxing
- OpenAI: Codex CLI on GitHub
- Anthropic: Claude Code overview, setup and plans and MCP
- Anthropic: Skills, CLAUDE.md memory and permission modes
Checked on September 28, 2026. OpenAI's Codex pages at developers.openai.com now redirect to learn.chatgpt.com. Both products change often, so confirm plans and commands at the source.