Frequently asked questions about Cursor, Claude Code, and Codex cover the differences between the three tools, model choice, pricing, whether vendors train on customer code, and the daily mechanics of each: Cursor modes and Tab, Claude Code install and permissions, Codex surfaces and approval modes, plus MCP servers, Agent Skills, and cost control.
You installed the tool, opened a project, and now you have questions. This page answers the ones developers actually ask — from “which model should I use?” to “why did my included usage run out sooner than expected?”
No. Line-by-line review cannot keep up once agents write most of the diff. Give the session a feedback loop, run layered agent review (REVIEW.md, Bugbot, claude-code-action, or Codex review), and keep a human on intent, risk, merge, and production. Read every line only for regulated or critical code, or when a finding says the change can break behavior, leak data, or breach policy. See the AI-native SDLC and Human in the loop.
What is the difference between Cursor, Claude Code, and Codex?
Cursor is an AI-native IDE (fork of VS Code) with inline editing, tab completions, and agent mode built in. Claude Code is a CLI-first agent that also spans IDE, desktop, browser, and Slack. Codex is OpenAI’s multi-surface agent inside the ChatGPT desktop app, CLI, IDE extension, and cloud service with GitHub/Slack/Linear integrations.
Pick Cursor if you want an editor-first Agents Window and Tab workflow. Pick Claude Code for Anthropic’s agent across terminal, IDE, desktop, browser, and Routines. Pick Codex for its shared ChatGPT desktop, CLI, IDE, and Cloud workflow; isolated desktop tasks can use worktrees when selected.
For complex coding tasks (default): Claude Opus 5, GPT-5.6 Terra, or Grok 4.6. Escalate planning to Claude Fable 5 or GPT-5.6 Sol. Use Sonnet 5, Haiku 4.5, GPT-5.6 Luna, Gemini 3.8 Flash, or Muse Spark 1.3 for volume and subagents.
For everyday work on a budget: Claude Sonnet 5 ($2 / $10 per MTok) or GPT-5.6 Luna ($1 / $6). Gemini 3.8 Flash is $0.75 / $3.75 introductory through 2026-12-31.
For Codex users: GPT-5.6 Terra as the daily default; Sol for hard planning; Luna for volume. All three share a 1M-token context window.
For speed-critical work in Cursor: Grok 4.6 and Composer 2.5 sit in the Cursor Models pool (included usage). Third-party flagships draw from the Other Models pool.
For cheap long-horizon coding: Muse Spark 1.3 ($1.25 / $4.25) in Muse Code / Meta Model API.
Cursor: Free tier; Pro $20, Pro+ $60 ($70 third-party API usage included), and Ultra $200 ($400 third-party API usage included) per month. Agent usage is metered under each plan rather than split into Fast/Slow request buckets
Claude Code Pro: $20/month; Max 5x: $100/month; Max 20x: $200/month. All remain usage-limited and defaults vary by account/provider
Codex via ChatGPT Plus: $20/month
Codex via ChatGPT Pro 5x: $100/month; Pro 20x: $200/month. Included usage can be extended with token credits
Team Plans
Cursor Teams: $40/user/month
Claude Team / Enterprise: Seat- and contract-dependent
ChatGPT Business / Enterprise with Codex: Plan- and credit-dependent
Enterprise (all tools): Custom quotes
Claude Fable 5: Global access returned July 1. Since July 20, 2026 it is permanently included on Max and Team Premium, capped at 50% of weekly usage limits; Pro and Team Standard use usage credits instead.
Absolutely. Many developers use Cursor for interactive editing and tab completions, Claude Code for terminal-based autonomous tasks, and Codex for cloud-based automation and GitHub issue handling. The tools are complementary, not mutually exclusive. Your project configuration files (.cursor/rules/, CLAUDE.md, AGENTS.md) can coexist in the same repository.
All three tools offer business privacy controls, but terms depend on the product and plan:
Cursor: Privacy mode prevents code from being sent for training. SOC 2 certified.
Claude Code: Commercial/API terms exclude customer content from model training by default; consumer settings and retention can differ. Enterprise adds governance and audit controls.
Codex: API and business offerings do not use inputs for training by default; consumer ChatGPT data controls and retention depend on settings and plan.
Always verify the current privacy policy for your tier. Free tiers may have different terms.
Open the AI chat panel (Cmd/Ctrl+I), then click the mode selector dropdown at the top. Choose “Agent” for autonomous file editing or “Ask” for read-only exploration. You can also press Cmd/Ctrl+. to toggle modes.
Settings > Models > API Keys. Enter your Anthropic, OpenAI, or Google key, click Verify, then select “Use my API key” in the model dropdown. Own API keys work for chat models but not Tab completion or specialized features.
Cursor no longer documents the legacy Fast/Slow request buckets as the current plan model. Agent usage is metered against the allowance and rate card for your plan and selected model. Check the live breakdown in Settings > Usage; model price, Max Mode, and long contexts can consume included usage at different rates.
The old progressive token-budget ladder is gone. think, think hard, and think more are ordinary prompt text. ultrathink is the one current exception: Claude Code recognizes it anywhere in a prompt and adds a deeper-reasoning instruction for that single turn, but it does not change the API effort level or reserve a fixed token budget. Use the thinking toggle and /model or /effort for durable control.
Claude Code limits depend on plan, selected model, effort, context size, and tool use; Max plans raise included usage but are not unlimited. Long conversations with big files consume capacity faster. Check the live usage view rather than estimating a fixed message count. Strategies:
Use /compact to summarize long conversations
Use /clear between unrelated tasks
Batch related questions into single prompts
Upgrade to Max 5x ($100/month) or Max 20x ($200/month)
Use usage credits or API access for extremely heavy workloads
Yes. Install the “Claude Code” extension from the VS Code marketplace for an integrated agent experience with editor context, diffs, and the same underlying Claude Code session — not merely an embedded terminal.
Worktrees are an isolation option for local Codex tasks in the ChatGPT desktop app, not a universal property of every surface. When you choose an isolated desktop task, Codex uses a separate checkout so:
Multiple tasks can run in parallel without file conflicts
Each task has its own clean working directory
Completed tasks produce a diff that you review before merging
Failed tasks are discarded without affecting your main branch
CLI sessions normally operate in the working directory you launch them from, while Cloud uses its own remote environment. For an isolated desktop task, Codex manages the worktree lifecycle.
Approval policy and sandboxing are separate controls. Use on-request for interactive work so Codex can request an elevated action when needed. never means it will not ask: failures remain failures unless the configured sandbox already permits the operation. Pair never only with a deliberately chosen sandbox for trusted unattended jobs. The old --full-auto flag is a deprecated compatibility shortcut; configure approval and sandbox explicitly instead.
MCP (Model Context Protocol) servers extend what your AI agent can do by connecting it to external tools — databases, APIs, browsers, and more. You do not strictly need them, but they unlock powerful workflows. For example, the GitHub MCP lets the agent create PRs directly, and the Postgres MCP lets it query your database.
All three tools support MCP. Start with one or two essential servers and add more as needed.
MCP servers maintain a persistent connection and expose multiple tools. They are heavier to set up but more powerful for deep integrations (databases, browser automation, cloud services).
Agent Skills are lightweight, single-purpose augmentations installed via npx skills add <owner/repo>. They work across 35+ agents and are easier to share. Think of Skills as focused recipes and MCP servers as full integration platforms.
When both exist for a workflow, Skills are faster to set up; MCP servers offer deeper, persistent access.
Or add a [mcp_servers.github] table to ~/.codex/config.toml (project-scoped .codex/config.toml for trusted projects). The Codex IDE extension exposes this via MCP settings > Open config.toml.
A systematic approach for building complex features with AI:
PRD: Write a clear specification with user stories and acceptance criteria
Plan: Ask the AI to create a detailed implementation plan. In Claude Code, raise /effort or use the thinking toggle for complex plans; think hard is ordinary prompt text
Todo: Convert the plan into a checklist. The AI works through items systematically, checking them off as it goes
This works identically across all three tools. The key is giving the AI a structured starting point rather than a vague request.
Batch operations: Group related changes into a single conversation
Frequently asked questions
What is the difference between Cursor, Claude Code, and Codex?
Cursor is an AI-native IDE (a VS Code fork) with inline editing, Tab completions, and agent mode built in. Claude Code is Anthropic's CLI-first agent that also spans IDE, desktop, browser, and Slack. Codex is OpenAI's multi-surface agent across the ChatGPT desktop app, CLI, IDE extension, and cloud service with GitHub, Slack, and Linear integrations.
Which AI model should I use?
Claude Fable 5 for maximum Claude capability; Claude Code's default depends on plan (Sonnet 5 on Pro and Team Standard, Opus 5 on Max and Team Premium); GPT-5.6 Sol, Terra, or Luna for Codex; Grok 4.5 for long-running Cursor work; Cursor Composer 2.5 for speed-critical loops; Gemini 3.1 Pro for extreme context or multimodal work.
How much does it cost?
Cursor has a free tier plus Pro $20, Pro+ $60, and Ultra $200 per month. Claude Code Pro is $20/month, Max 5x $100/month, Max 20x $200/month. Codex is included with ChatGPT Plus ($20/month) or Pro (from $100/month), extendable with token credits. Cursor Teams is $40/user/month; Claude and ChatGPT team pricing is seat- and contract-dependent.
Can I use multiple tools together?
Yes — they are complementary. Many developers use Cursor for interactive editing, Claude Code for terminal-based autonomous tasks, and Codex for cloud automation and GitHub issue handling. Their configuration files (.cursor/rules/, CLAUDE.md, AGENTS.md) can coexist in one repository.
Do these tools train on my code?
All three offer business privacy controls. Cursor's privacy mode prevents code from being sent for training (SOC 2 certified); Claude's commercial/API terms exclude customer content from training by default; Codex API and business offerings do not use inputs for training by default. Consumer tiers differ — verify the current policy for your plan.
How do I switch between Agent and Ask mode in Cursor?
Open the AI chat panel with Cmd/Ctrl+I and pick the mode in the dropdown at the top, or press Cmd/Ctrl+. to toggle. Use Ask mode to explore and plan first, then Agent mode to execute.
Why is Cursor's Tab autocomplete not working?
Check your plan's live allowance in Settings > Usage, verify codebase indexing shows "Indexed", check status.cursor.sh for outages, confirm the file type is supported, and restart Cursor if state is corrupted.
How do I use my own API key in Cursor?
Settings > Models > API Keys: enter an Anthropic, OpenAI, or Google key, click Verify, then select "Use my API key" in the model dropdown. Own keys work for chat models but not Tab completion or specialized features.
How is Cursor agent usage metered?
Against your plan's allowance and rate card rather than the legacy Fast/Slow request buckets. Model price, Max Mode, and long contexts consume included usage at different rates — Settings > Usage shows the live breakdown.
How do I set up BugBot for PR reviews?
Install the BugBot GitHub App from your Cursor team dashboard and create .cursor/bugbot.yml in the repository. BugBot then reviews new PRs and posts inline comments, each with a "Fix in Cursor" action.
How do I install Claude Code?
Run the native installer (curl -fsSL https://claude.ai/install.sh | bash) or, on macOS, brew install --cask claude-code. Then run "claude" in any project directory and authenticate on first run.
What is the difference between think, think hard, and ultrathink in Claude Code?
The old progressive token-budget ladder is gone — "think" and "think hard" are ordinary prompt text. "ultrathink" is the one exception: Claude Code recognizes it anywhere in a prompt and adds a deeper-reasoning instruction for that single turn. Use the thinking toggle and /model or /effort for durable control.
How do I stop getting permission prompts in Claude Code?
Pre-approve specific operations through Claude Code's permission system, or — only in sandboxed environments such as Docker containers or CI — run claude --dangerously-skip-permissions.
Why did I hit my Claude Code message limit so quickly?
Limits depend on plan, selected model, effort, context size, and tool use — long conversations with big files consume capacity faster. Use /compact to summarize, /clear between unrelated tasks, batch related questions, or upgrade to Max 5x ($100/month) or Max 20x ($200/month).
Can I use Claude Code inside VS Code?
Yes. Install the "Claude Code" extension from the VS Code marketplace for an integrated agent experience with editor context, diffs, and the same underlying Claude Code session.
What are the different Codex surfaces?
Four: the ChatGPT desktop app (local tasks with cloud handoff), the open-source Codex CLI, the Codex IDE extension, and Codex Cloud — background agents triggered from ChatGPT, GitHub, Slack, or Linear.
How do worktrees work in Codex?
For isolated desktop tasks, Codex uses a separate checkout so parallel tasks don't conflict; each task gets a clean working directory and produces a diff you review before merging, and failed tasks are discarded without affecting your main branch.
How do I set up Codex automations?
Assign a GitHub issue to @codex or mention it in a PR comment, message Codex in a configured Slack channel, assign a Linear ticket to the Codex integration, or configure scheduled automations in the ChatGPT desktop app. Each spawns a cloud agent that posts results or a PR.
What are Codex's approval modes?
Approval policy and sandboxing are separate controls. Use on-request for interactive work so Codex can ask to elevate when needed; "never" means it will not ask, so pair it only with a deliberately chosen sandbox for trusted unattended jobs.
What are MCP servers and do I need them?
MCP (Model Context Protocol) servers connect your agent to external tools — databases, APIs, browsers. You don't strictly need them, but they unlock workflows like creating PRs via the GitHub MCP or querying your database via the Postgres MCP. All three tools support MCP.
How do MCP servers differ from Agent Skills?
MCP servers maintain a persistent connection and expose multiple tools — heavier to set up, deeper integrations. Agent Skills are lightweight, single-purpose augmentations installed via "npx skills add <owner/repo>" that work across 35+ agents and are easier to share.
How do I install an MCP server?
Cursor: Settings > Features > MCP > "+ Add MCP Server". Claude Code: "claude mcp add github -s user -- npx -y @modelcontextprotocol/server-github" or edit .mcp.json. Codex: "codex mcp add github -- npx -y @modelcontextprotocol/server-github" or a [mcp_servers] table in config.toml.
What is the PRD-to-Plan-to-Todo methodology?
Write a clear PRD with user stories and acceptance criteria, ask the AI to create a detailed implementation plan, then convert the plan into a checklist the AI works through systematically. It works identically across all three tools.
Should I use Agent or Chat/Ask mode?
Agent mode for multi-file changes, implementing features from a plan, refactoring, and running commands. Ask/Chat mode for exploring and understanding code, planning before implementing, and reviewing options before committing to an approach.
How do I optimize AI coding tools for cost?
Choose the right model for the task, clear context regularly, write specific prompts, reference only the files the AI needs, monitor usage (Cursor Settings > Usage, /cost in Claude Code, the Codex dashboard), and batch related changes into a single conversation.