Zed vs Cursor: open source native editor or AI-first VS Code fork

What is the difference between Zed and Cursor?

Zed is an open source code editor with built-in AI that you can run free with your own API keys or outside agents, while Cursor is an AI-first editor based on the VS Code codebase that bundles its own agent and model usage into a paid subscription. Choose Zed for an open, bring-your-own-model setup; choose Cursor for a VS Code workflow with AI built in end to end.

Both have a coding agent in a side panel, AI autocomplete, AGENTS.md support, Agent Skills and MCP. The real differences are the editor underneath, who supplies the models and how you pay. Everything here comes from the official docs as read on 26 October 2026. For deeper single-tool guides, see Zed AI and Cursor Agent.

How does Zed work?

Zed's GitHub repository calls it "a high-performance, multiplayer code editor from the creators of Atom and Tree-sitter." The README says the source is "licensed primarily under GPL-3.0-or-later, with Apache-2.0 components where marked," and that you can download Zed on macOS, Linux and Windows.

Zed gives you three ways to run an agent:

• Zed Agent, the native agent, which can read and search your project, edit files, run terminal commands, use MCP tools and show changes in Zed's review UI, per the Zed Agent page.
• External Agents, installed from the ACP Registry. The external agents page lists Claude Agent, Codex, Gemini CLI, OpenCode, Copilot, Cursor, Pi Coding Agent and Poolside, and says "Zed does not charge for External Agents."
• Terminal Threads, where any agent CLI runs in a terminal that Zed tracks as a thread.

Parallel Agents "lets you run multiple agent threads and Terminal Threads at once," and you can start one in a new Git worktree when two threads might edit the same files.

How does Cursor work?

Cursor's VS Code migration page says "Cursor is based upon the VS Code codebase" and that Cursor is regularly rebased onto the latest VS Code version. A one-click import under General > Account brings over extensions, themes, settings and keybindings.

The Agent overview says "Agent is Cursor's assistant that can complete complex coding tasks independently, run terminal commands, and edit code." You open it in the side pane with Cmd+I. Shift+Tab cycles through its modes: Agent, read-only Ask, Plan, where Agent asks clarifying questions and writes a plan you can edit before it builds, and Debug, which uses runtime logs to find a bug's cause. Checkpoints save snapshots during a session so you can restore files.

Beyond the editor, Cursor has cloud agents that "run in isolated VMs in the cloud with full development environments," and a CLI that runs the same agent in a terminal.

How do Zed and Cursor compare side by side?

The Claude Code row for Cursor comes from the Claude Code docs, which offer an "Install for Cursor" link for the extension.

How do the agents handle approvals?

Zed asks by default. The tool permissions page documents agent.tool_permissions.default as "confirm" by default, and you can set per-tool defaults for terminal commands and file edits plus regex patterns with always_allow, always_confirm and always_deny.

Cursor uses Run Modes under Settings > Agents > Approvals & Execution. The run modes page lists three:

• Auto-review, which Cursor calls "the safest useful setup for most people," runs allowlisted calls, sandboxes shell commands when it can and sends other calls to a classifier.
• Allowlist runs only actions on your allowlist without approval.
• Run Everything runs every tool call automatically, with no sandbox or classifier.

On macOS, Cursor's sandbox "uses Seatbelt through sandbox-exec," and on Linux it uses Landlock and seccomp.

Cursor's Agent security page says that by default terminal commands need your approval, while its run modes changelog says Auto-review shipped as the recommended default in version 3.6, so check the setting on your install. Zed starts at "ask me every time." Both let you move in either direction.

How do rules, skills and MCP compare?

Rules. The Zed instructions page says "Zed supports AGENTS.md as the primary instruction file," but it uses only the first file it finds in this order: .rules, .cursorrules, .windsurfrules, .clinerules, .github/copilot-instructions.md, AGENT.md, AGENTS.md, CLAUDE.md, GEMINI.md. A repo with an old .cursorrules file gives Zed that file, not your AGENTS.md. Cursor's rules docs say project rules "live in .cursor/rules as .mdc files," and AGENTS.md works in the root and in subdirectories, with nested files taking precedence. The Cursor rules guide and AGENTS.md guide go further.

Skills. Zed reads skills only from ~/.agents/skills/ and .agents/skills/, per its skills page, and suggests a symlink for other folders. Cursor's skills docs list .agents/skills/ and .cursor/skills/, user-level equivalents, and compatibility folders .claude/skills/ and .codex/skills/. A shared .agents/skills/ folder works in both. See Cursor skills and Agent Skills.

MCP. Zed stores servers under context_servers and supports MCP's "Tools and Prompts features." It also forwards them to external agents over ACP. Cursor uses .cursor/mcp.json per project or ~/.cursor/mcp.json globally, with stdio, SSE and Streamable HTTP transports. The Cursor MCP guide has examples.

How do autocomplete and pricing compare?

Autocomplete. Zed calls it edit prediction. The default provider is Zeta, "an open source model developed by Zed," included with Pro, the Pro trial, and the Student and Business plans; on the free plan you can use GitHub Copilot, Mercury Coder, Codestral or a local model, per the edit prediction docs. Cursor's docs describe Tab as "Cursor's AI-powered autocomplete," covered in the Cursor Tab guide.

Pricing, as of October 2026.

Zed Pro bills hosted usage at "API list price +10%," and there is a 14-day trial with no credit card. Cursor's page says "Third Party model usage in Cursor is not included in the monthly prices shown and must be purchased separately," and that Premium and higher include third-party models at API list price. Cursor subscriptions are now sold through X Subscription Corp. The Cursor pricing guide breaks down the usage pools.

Which should you pick?

Start with whether you want to bring your own models or buy a bundle.

Zed tends to fit when:

• You want an open source editor and are happy without VS Code extensions.
• You already pay for Claude, ChatGPT or another agent and want to run it in your editor at no extra charge from Zed.
• You want approvals on by default and a multiplayer editor.
• You want to run several agents, including Cursor's, side by side.

Cursor tends to fit when:

• You live in VS Code and want your extensions, keybindings and settings to carry over.
• You want one subscription that covers the agent, autocomplete and cloud agents.
• You want a sandbox and classifier deciding which commands run without asking.
• You use skills from .claude/skills/ or .codex/skills/ and want them picked up without changes.

You do not have to commit to one setup forever. Keep instructions in AGENTS.md and skills in .agents/skills/, and both editors read them. If Claude Code is in the mix too, see Claude Code vs Cursor.

Keep your prompts and skills portable

Editors change; your best prompts and skills should not have to. Kipt keeps prompts and Agent Skills as Markdown in a private GitHub repo you own, with {{name}} blanks and Cmd+K search, and can import skills from a GitHub link, a folder or a .zip. If a prompt you reuse belongs in .agents/skills/, the free SKILL.md generator turns it into a skill folder in your browser.