Claude Code vs Cursor vs Codex: how do they compare?
Short answer DER
All 3 document both MCP and Agent Skills support. Claude Code is a coding agent from Anthropic, Cursor is an AI code editor from Anysphere and Codex is a coding agent from OpenAI. Codex has open-source code on record (Apache-2.0); the others are proprietary. As of 2026-11-21, publishers name them in 1,377 (Claude Code), 772 (Cursor) and 788 (Codex) catalogue listings: vendor claims, not tested compatibility.
The evidence
Claude Code vs Cursor vs Codex side by sideCUR
Curated from each vendor's documentation, pricing and repository pages. "Listings naming it" counts catalogue products whose publisher names the platform (vendor claims).
| Claude Code | Cursor | Codex | |
|---|---|---|---|
| Kind | coding agent | AI code editor | coding agent |
| Vendor | Anthropic | Anysphere | OpenAI |
| What it is | Anthropic's agentic coding tool that reads a codebase, edits files, runs commands and manages git, available as a terminal CLI, VS Code and JetBrains extensions, a desktop app, and cloud sessions on the web and mobile. | AI code editor (a VS Code fork) with an in-editor coding agent, plus a headless CLI, cloud agents reachable from web, mobile and Slack, Bugbot PR review, and an agent SDK. | OpenAI's software engineering agent that reads, edits and runs code locally (CLI, IDE extension, ChatGPT desktop app) or in isolated cloud workspaces (Codex Cloud), and can be driven from GitHub, Slack, Linear or an SDK. |
| Open source | No | No | Yes |
| Licence | None on record | None on record | Apache-2.0 |
| Pricing model | paid | freemium | Unknown |
| Extension mechanisms | mcp, agent-skills, plugins, hooks, subagents, slash-commands, rules, workflows | rules, mcp, agent-skills, hooks, subagents, plugins, slash-commands | mcp, agent-skills, plugins, hooks, subagents, slash-commands, rules |
| MCP | Documented | Documented | Documented |
| Agent Skills | Documented | Documented | Documented |
| Surfaces | cli, ide, desktop, web, mobile, slack, api | ide, desktop, cli, web, mobile, slack, api | cli, ide, desktop, web, mobile, slack, api |
| Listings naming it | 1,377 | 772 | 788 |
| Status | Active | Active | Active |
| Notes | The public GitHub repo is for issues/plugins and has no OSI license; the product is not open source. Included in Pro, Max, Team and Enterprise plans; also usable with Anthropic Console API keys or third-party providers (Bedrock, Vertex, etc.). 'rules' = CLAUDE.md / AGENTS.md memory files. Agent SDK provides the 'api' surface. Official plugin marketplace is claude-plugins-official. | Plans: Free, Plus ($8), Premium ($30), Super ($100), Ultra ($200), Teams ($30+/user), Enterprise. Docs reference a plugin marketplace (marketplace-security help page) but no public marketplace URL was confirmed. Cursor acquired Continue in 2026 (see continue). | Only the Codex CLI (openai/codex) is open source; the cloud and app surfaces are proprietary. developers.openai.com/codex now redirects to learn.chatgpt.com/docs. Codex app for macOS launched Feb 2, 2026; Codex Cloud launched Sep 29, 2026; plugins in Codex Mar 26, 2026. Access is tied to ChatGPT plans, but the plan matrix was not verified, so pricing is left null. |
| Checked on | 2026-11-21 | 2026-11-21 | 2026-11-21 |
How this comparison was made. Each platform record was researched by AgentSidekick against the official sources listed below and dated in "Checked on". Fields we could not confirm are "Unknown", not guessed.
Limit. AgentSidekick has not benchmarked code quality, speed or cost. Pricing model is a category; plan prices change often and are only in the notes where we confirmed them.
Which surfaces do they share? All 3 are available on: cli, ide, desktop, web, mobile, slack, api.
Which extension mechanisms are unique to one of them? Claude Code: workflows.
Which has the most MCP servers naming it? Claude Code 507, Cursor 455, Codex 370 active MCP servers name each in their descriptions. See best MCP servers for Claude Code.
Related
Sources
Pages the records behind this answer were taken from. Counts and rankings are computed by the site from these records.
- Claude Code: code.claude.com (external)checked
- Claude Code: code.claude.com (external)checked
- Claude Code: claude.com (external)checked
- Claude Code: github.com (external)checked
- Cursor: cursor.com (external)checked
- Cursor: cursor.com (external)checked
- Cursor: cursor.com (external)checked
- Cursor: github.com (external)checked
- Codex: developers.openai.com (external)checked
- Codex: learn.chatgpt.com (external)checked
- Codex: learn.chatgpt.com (external)checked
- Codex: github.com (external)checked
- Codex: help.openai.com (external)checked
How claims are labelled and ranked: claim kinds, ranking rules, compatibility statuses. Spotted an error? See corrections.