Compare · Comparison
Claude Code vs Codex vs GitHub Copilot vs Cursor
Four coding agents compared on plans, where they run, file access and admin controls. For people who work in code. Sourced from official pages.
What stays yours: Reviewing code and approving changes before they merge.
Which level of help each one gives: Assist, Augment, Adapt explained.
Not sure which kind of tool you need? Choosing how a tool works with you.
| Field | Claude Code | Codex | GitHub Copilot | Cursor |
|---|---|---|---|---|
| How it works with you | It works in your files | It works in your files | It works in your files | It works in your files |
| How much it does on its own | Works on its own | Works on its own | Works on its own | Works on its own |
| Which plans include it, and cost | All paid Claude plans (Pro, Max, Team, Enterprise); shares plan usage pool | Included on all ChatGPT plans, including Free and Go; usage limits vary by plan. Codex Cloud on eligible Plus, Pro, Business, Enterprise, Healthcare and Education plans (not Free or Go) | Free (limited), Pro $10/mo, Pro+ $39/mo, Max $100/mo; usage in GitHub AI Credits ($0.01 each) | Free (limited), Plus $8/mo (Grok models only), Premium $30/mo, Super $100/mo, Ultra $200/mo, Teams $30+/user/mo; third-party models at API list price |
| Where you can use it | terminal, desktop app, VS Code, JetBrains | ChatGPT desktop app (Codex mode), Codex CLI, Codex IDE extension, Codex web | VS Code, JetBrains, CLI, github.com, GitHub Mobile | desktop editor, cloud agents (Premium+) |
| What it can connect to | MCP supported | MCP servers (local STDIO and streamable HTTP) in the ChatGPT desktop app, Codex CLI and IDE extension, sharing one configuration; plugins | Cloud agent integrates with MCP servers | MCP access controls on Enterprise |
| Can it open and edit your files? | Works directly on files in a workspace | Local tasks work on your device in your project; cloud tasks run on OpenAI-managed computers in reusable cloud environments | Repos/files | Edits project files |
| Controls for your organisation | Not yet confirmed | Managed workspaces control Codex Local and Codex Cloud separately; role-based access; Compliance API logs; Enterprise cloud access off by default | Business/Enterprise | Teams/Enterprise |
- How it works with you
- It works in your files
- How much it does on its own
- Works on its own
- Which plans include it, and cost
- All paid Claude plans (Pro, Max, Team, Enterprise); shares plan usage pool
- Where you can use it
- terminal, desktop app, VS Code, JetBrains
- What it can connect to
- MCP supported
- Can it open and edit your files?
- Works directly on files in a workspace
- Controls for your organisation
- Not yet confirmed
- How it works with you
- It works in your files
- How much it does on its own
- Works on its own
- Which plans include it, and cost
- Included on all ChatGPT plans, including Free and Go; usage limits vary by plan. Codex Cloud on eligible Plus, Pro, Business, Enterprise, Healthcare and Education plans (not Free or Go)
- Where you can use it
- ChatGPT desktop app (Codex mode), Codex CLI, Codex IDE extension, Codex web
- What it can connect to
- MCP servers (local STDIO and streamable HTTP) in the ChatGPT desktop app, Codex CLI and IDE extension, sharing one configuration; plugins
- Can it open and edit your files?
- Local tasks work on your device in your project; cloud tasks run on OpenAI-managed computers in reusable cloud environments
- Controls for your organisation
- Managed workspaces control Codex Local and Codex Cloud separately; role-based access; Compliance API logs; Enterprise cloud access off by default
- How it works with you
- It works in your files
- How much it does on its own
- Works on its own
- Which plans include it, and cost
- Free (limited), Pro $10/mo, Pro+ $39/mo, Max $100/mo; usage in GitHub AI Credits ($0.01 each)
- Where you can use it
- VS Code, JetBrains, CLI, github.com, GitHub Mobile
- What it can connect to
- Cloud agent integrates with MCP servers
- Can it open and edit your files?
- Repos/files
- Controls for your organisation
- Business/Enterprise
- How it works with you
- It works in your files
- How much it does on its own
- Works on its own
- Which plans include it, and cost
- Free (limited), Plus $8/mo (Grok models only), Premium $30/mo, Super $100/mo, Ultra $200/mo, Teams $30+/user/mo; third-party models at API list price
- Where you can use it
- desktop editor, cloud agents (Premium+)
- What it can connect to
- MCP access controls on Enterprise
- Can it open and edit your files?
- Edits project files
- Controls for your organisation
- Teams/Enterprise
As of . Sources: claude.com/pricing · claude.com/product/claude-code · chatgpt.com/pricing · help.openai.com/en/articles/11369540-using-codex-with-your-chatgpt-plan · learn.chatgpt.com/docs/extend/mcp · github.com/features/copilot/plans · cursor.com/pricing
For people who work in code.
TL;DR
All four are Code harnesses: Claude Code, Codex, GitHub Copilot and Cursor. Pick by plan fit, where the agent runs, how it reaches your files, and who must approve the pull request. Craft ratings place each in Adapt; your role is still to set limits and review what lands in the repo.
Facts below come from official pricing and product pages. Where a field is not yet confirmed, the table shows that gap.
How the four line up
Claude Code (Anthropic) is included on all paid Claude plans (Pro, Max, Team, Enterprise) and shares the plan usage pool. It runs in the terminal, a desktop app, VS Code and JetBrains. MCP is supported. It works directly on files in a workspace. Organisational limits come through the Claude plan you use. Your job: set the rules and review changes before they ship.
Codex (OpenAI) is included across ChatGPT plans, including Free and Go, with usage limits that vary by plan. Codex Cloud is available to eligible Plus, Pro, Business, Enterprise, Healthcare and Education accounts, and is not included with Free or Go. It runs in the ChatGPT desktop app (Codex mode), the Codex CLI, the Codex IDE extension and Codex web. MCP servers (local STDIO and streamable HTTP) work in the desktop app, CLI and IDE extension with one shared configuration, plus plugins. Local tasks run on your device in your project; cloud tasks run on OpenAI-managed computers in reusable environments. Managed workspaces control Codex Local and Codex Cloud separately, with role-based access and Compliance API logs; Enterprise cloud access is off by default. Your job: set task limits and approve code before it merges.
GitHub Copilot (GitHub) lists Free (limited), Pro at $10/mo, Pro+ at $39/mo and Max at $100/mo, with usage in GitHub AI Credits at $0.01 each. It runs in VS Code, JetBrains, the CLI, github.com and GitHub Mobile. The cloud agent integrates with MCP servers. File access is repos and files. Admin controls sit on Business and Enterprise. Your job: approve pull requests before anything merges.
Cursor lists Free (limited), Plus at $8/mo (Grok models only), Premium at $30/mo, Super at $100/mo, Ultra at $200/mo and Teams from $30+/user/mo, with third-party models at API list price. Platforms: desktop editor, plus cloud agents on Premium and above. MCP access controls are on Enterprise. It edits project files. Scheduling of cloud agents and automations is on Premium+. Admin controls are Teams and Enterprise. Your job: review and accept every change to your code.
Plans and cost shape
Claude Code has no separate seat fee beyond the paid Claude plan. Codex is billed through the ChatGPT plan ladder; OpenAI does not list a separate dollar price for Codex itself on the pages we checked. Copilot Credits and Cursor’s third-party API list price can dominate the headline fee—measure a representative task before you budget a rollout.
For a team decision, run the same small coding task in the surfaces you already pay for. Record completion quality, review time and billable usage. The plans row in the table is the starting map; your invoice is the verification.
Where it runs and what it can touch
Claude Code’s listed surfaces are terminal, desktop, VS Code and JetBrains, with direct workspace file access and MCP support. That combination suits people who already live in an editor or CLI and want the agent on the same tree they will merge.
Codex spans the ChatGPT desktop app, CLI, IDE extension and web. Local work stays on your device; cloud work runs in OpenAI-managed environments and can continue while your computer is asleep.
GitHub Copilot spans editors, CLI, github.com and GitHub Mobile, with repo/file access and MCP on the cloud agent. It is the natural trial when the review culture already centres on GitHub pull requests and Business/Enterprise admin.
Cursor is a desktop editor first, with cloud agents on Premium+. File edits are to project files; MCP controls appear on Enterprise. Use it when the editor itself is the product you evaluate, not only an extension inside another IDE.
Admin controls and the human gate
GitHub Copilot and Cursor both list organisational admin on Business/Enterprise or Teams/Enterprise. Claude Code’s organisational controls live on the Claude Team and Enterprise side of the account. Codex lets managed workspaces control local and cloud use separately, with role-based access and Compliance API logs, and keeps Enterprise cloud access off until enabled.
Across all four, the Craft human role is the same shape: someone who can read the diff must approve before merge. More autonomy in the harness does not remove that gate.
Craft mode
The tracker rates all four Adapt: the agent works on code within rules you set; you stay on call. That is editorial, not a vendor claim that Assist or Augment never apply. Use Assist when you are still clarifying the design; Augment when the agent drafts several steps you will inspect; Adapt when it edits the workspace under explicit stop rules.
For Anthropic’s chat and cowork surfaces beside Code, see Claude Chat vs Cowork vs Claude Code.
Who should trial which first
Already on a paid Claude plan: trial Claude Code in the editor or terminal you use daily. Confirm MCP and workspace scope against a throwaway branch.
Already on ChatGPT: trial Codex on the plan you have. Free and Go include Codex with plan usage limits; Plus and above can add Codex Cloud where eligible. Confirm local versus cloud behaviour against a small task before a team mandate.
Living in GitHub PRs: trial Copilot on a Free or Pro seat, measure AI Credit use, and involve Business/Enterprise admins early if policy matters.
Want an AI-first editor: trial Cursor’s Free or Plus tier (Plus is Grok models only), then Premium if you need cloud agents—review every accepted edit.
FAQ
Which coding agent should I start with?
Start from the vendor and surfaces you already pay for. Claude Code sits on paid Claude plans; Codex on ChatGPT plans (with Codex Cloud on eligible paid tiers); GitHub Copilot on Free through Max with credit metering; Cursor on Free through Ultra and Teams. Then compare file access and admin needs for the repo you will actually change.
Do all four support MCP?
Claude Code lists MCP as supported. Codex supports MCP servers in the ChatGPT desktop app, Codex CLI and IDE extension, sharing one configuration. GitHub Copilot’s cloud agent integrates with MCP servers. Cursor lists MCP access controls on Enterprise.
How do the entry prices compare?
GitHub Copilot Pro lists $10/mo; Cursor Plus lists $8/mo (Grok models only) and Premium $30/mo; Claude Code comes with paid Claude plans (Pro entry $20/mo monthly on Claude’s pricing page). Codex follows the ChatGPT plan ladder without a separate dollar line on the pricing page we checked. Usage meters (Copilot Credits, Cursor third-party API list price, Claude shared usage pool, ChatGPT plan limits) can dominate the headline fee.
What still stays human?
Reviewing the diff, approving the pull request and owning merge risk. The “what stays yours” lines for these tools all point at limits, approval and review before code ships—not at unattended production changes.
Before you rely on this
- Craft Adapt ratings are editorial.
- Prices are US-page figures from official pages as of 17 Oct 2026.
- A separate dollar price for Codex beyond the ChatGPT plan ladder is not shown on the pages we checked.