Cursor - Tool details — Yiding Huiziyou

EDITOR NOTE · Editorial advice, based on public sources

Cursor is for developers with a real repository who want Agent assistance in file context. Editor, models, MCP, Cloud Agents, and Bugbot share one workflow, but external-model usage follows token/API pricing and can exceed the subscription.

Checked: 2026-10-12 · Official site material

Best for

Focused refactors, debugging, test-driven edits, code review, and multi-file Agent work in an existing project.

Not for

No-code product prototyping, or teams needing a strictly fixed budget without monitoring model usage.

Key capabilities

In-editor Agent

Official docs center Agent, Chat, Tab completions, and Composer. Hobby has limited Agent use; paid plans extend Agent access and support frontier models.[1][3]

Usage pools by model

Pro, Pro+, and Ultra include Cursor Models and Other Models pools. Third-party models consume usage at provider API prices, so model selection changes burn rate.[3][2]

MCP, Cloud Agents, and Bugbot

Paid individual tiers list MCP, skills, hooks, Cloud Agents, and usage-billed Bugbot; team tiers add privacy enforcement, administration, SSO, and shared context.[1][2]

Pricing and free allowance

Hobby: Free; No card required; limited Agent requests and Composer. The help page also describes limited Agent, Chat, and Tab use.[1][2]

Pro / Pro+ / Ultra: USD 20 / USD 60 / USD 200 per month; annual billing page says about 20% savings, with exact monthly equivalent at checkout; The official models page lists about $20, $70, and $400 of Other Models usage pools for Pro, Pro+, and Ultra, plus bonus usage; extra use can be purchased at cost.[2][3]

Teams: USD 40/user/month; Premium USD 120/user/month, Enterprise custom; Team pricing adds centralized billing, privacy mode, administration, and SAML/OIDC SSO; current docs define the team Agent-request and pooling rules.[1][2]

Plan	Cost	Includes and terms
Hobby	Free	No card required; limited Agent requests and Composer. The help page also describes limited Agent, Chat, and Tab use.[1][2]
Pro / Pro+ / Ultra	USD 20 / USD 60 / USD 200 per month; annual billing page says about 20% savings, with exact monthly equivalent at checkout	The official models page lists about $20, $70, and $400 of Other Models usage pools for Pro, Pro+, and Ultra, plus bonus usage; extra use can be purchased at cost.[2][3]
Teams	USD 40/user/month; Premium USD 120/user/month, Enterprise custom	Team pricing adds centralized billing, privacy mode, administration, and SAML/OIDC SSO; current docs define the team Agent-request and pooling rules.[1][2]

Limitations to watch

Complex requests burn faster

Usage follows token/API prices; long context, Max Mode, and expensive models burn allowance faster, so request counts alone are unreliable.[3]

Overage can add cost

After included usage is exhausted, the editor offers additional usage or an upgrade; set spend limits and monitor the dashboard.[3][1]

AI edits still need code review

Agent, Background Agent, and Bugbot improve speed, but generated diffs, dependencies, and fixes need local tests and review before merging.[1][3]

Editorial workflow: how to try it

Fix one bug in an existing repo

Give Agent the issue, reproduction command, and relevant files; ask for the root cause before changing only necessary files.

Review the diff before merge

Review the Agent diff line by line, require a regression test, and reject unrelated formatting or dependency upgrades.

Run checks and inspect usage

Run existing lint, typecheck, and tests, investigate failures, then inspect usage and model burn.

When is it worth paying?

Use Hobby for one real bug fix and observe Agent usage. Upgrade to Pro for sustained Agent work, frontier models, MCP, or Cloud Agents; consider Pro+ or Ultra only for daily parallel agents and set a spend cap first.

Alternatives in this library

Bolt.new

Choose Bolt.new for a shareable prototype without an existing repo; choose Cursor for precise diffs and local tests in an existing codebase.

Claude

Choose Claude when long-document reasoning matters more than in-editor code changes.

Sources

[1]Cursor Pricing2026-10-12
[2]Pricing and plans2026-10-12
[3]Models & Pricing2026-10-12

Research note: Cursor's public pricing and model docs were read. Search results still contain older request-count pricing; current docs use model pools and API/token prices, so the older 500-request framing was not used.