AI tools·Coding agents·Global
Cursor
A default AI code editor. Tab, Composer, Agent, MCP, and model picking are the core.
Compiled through 2026-09
- Vendor
- Anysphere
- Started
- 2022
- China availability
- Overseas accessRequires overseas network access
- Pricing tier
- Free tier
- Current flagship
- Cursor Agent + multi-model (Claude 3.7 / GPT-5.6 / Grok 4.6)
- Website
- cursor.com
Evaluation, Pros & Cons
Editorial score based on public benchmarks and community feedback — not a third-party benchmark result
Key Advantages
- Gold standard AI Native IDE; Composer multi-file editing and codebase indexing
- Effortless model switching between Claude 3.7, GPT-4o, DeepSeek, plus first-class MCP support
Limitations & Caveats
- Monthly Pro subscription cost ($20/mo); fast-request limits apply on heavy usage
Overview
Cursor is a VS Code fork with Tab, chat, Composer, and repo-wide Agent. It supports MCP, rules, skills, Bugbot, and background agents. Models are interchangeable: OpenAI, Anthropic, Gemini, Grok, Kimi, and more. In 2026 GPT-5.6, Grok 4.5/4.6, and Claude Fable are first-class. AgentHub treats Cursor as a primary install target for MCP and Skills.
Direction
The direction is longer repo-level agents: background jobs, PR loops, tests, with MCP/Skills as the default tool layer. Model pickup will track each lab’s flagship. Cursor remains a subscribed editor, not a foundation-model lab.
Version history
Aug 2026
LatestGrok 4.6
Long-horizon coding model in Cursor.
View details
New capabilities
- Grok 4.6 for long-horizon coding
What got stronger
- Another cost-effective option for long agent runs
Version notes
With Grok 4.6 in Cursor, long jobs can pick among Claude Fable, GPT-5.6, and Grok. That is Cursor’s product logic.
Jul 2026
GPT-5.6 / Grok 4.5
Next-gen models in the picker.
View details
New capabilities
- GPT-5.6 and Grok 4.5 in the picker
What got stronger
- Same agent UX, newest flagship models
Version notes
Cursor does not train the base model; it tracks other labs’ flagships. The July 2026 drop is typical: same editor, new model pool.
Jun 2025
MCP & background agents
Tool protocol and async jobs.
View details
New capabilities
- MCP tool protocol
- Background / cloud agents for async jobs
What got stronger
- The editor can reach databases, browsers, and internal APIs without a one-off plugin
Version notes
MCP is why AgentHub treats Cursor as a primary client. Background agents made “it keeps working after I close the lid” a product feature.
Jan 2025
Agent mode
Plan-and-edit across the repo.
View details
New capabilities
- Agent mode: plan, edit the repo, run commands
What got stronger
- Jobs that span files and steps, not only the current buffer
Version notes
Agent mode made Cursor closer to “hand it the spec.” You still read the diff, but the unit of work is the repo, not the line.
Jun 2024
Composer
Multi-file edit becomes primary.
View details
New capabilities
- Composer edits many files at once
What got stronger
- From next-line complete to changing a whole feature
Version notes
Composer was the interaction that made Cursor mainstream: select, state intent, get a multi-file diff. Later Agent grew from this.
Mar 2023
Cursor public
VS Code-based AI editor.
View details
New capabilities
- VS Code-based AI editor
- Autocomplete plus chat-to-code
What got stronger
- More “the editor is the AI product” than a Copilot plugin
Version notes
Public Cursor treated AI as first-class, not a sidebar. Early model choice was thin; the interaction bet was already set.