Claude Code Cursor Extension: How to Install It and When to Use Novita AI

Yes, you can use Claude Code inside Cursor, but the important detail is that there is not a separate Cursor-only build. The right install path is the official Claude Code extension published by Anthropic, because Cursor can run many VS Code extensions. If your goal is the Claude Code sidebar workflow inside Cursor, install that extension. If your goal is broader model routing, lower token cost, or open-source coding models, run Claude Code from Cursor’s integrated terminal and point the backend at Novita AI.

Is There a Claude Code Cursor Extension?

Functionally, yes. Branding-wise, not exactly.

Anthropic documents Claude Code IDE integrations for VS Code forks, and Cursor fits that pattern. That means the install target is the official Claude Code extension, not a separate product called “Claude Code Cursor Extension.” If you search that exact phrase, you are really looking for one of two things:

1. The Anthropic extension running inside Cursor.
2. A Claude Code workflow that stays in Cursor while using a different API backend such as Novita AI.

Those are related, but they are not the same setup.

The first setup gives you the native Claude Code panel and Anthropic’s agent UX inside Cursor. The second setup is about backend control: you keep Cursor as your editor, but you run Claude Code from the integrated terminal and decide where model requests go.

How to Install Claude Code in Cursor

If what you want is the official extension experience, the setup is straightforward:

1. Open Cursor.
2. Open the Extensions panel.
3. Search for Claude Code.
4. Confirm the publisher is Anthropic.
5. Click Install.
6. Open the Claude sidebar from the activity bar.
7. Sign in when prompted.

That gets you the IDE integration path Anthropic officially supports.

In practice, this is the cleanest choice if:

• you want a sidebar-first agent workflow;
• you prefer reviewing diffs without leaving the editor;
• you already use Claude plans or Anthropic-hosted Claude models;
• you want the closest match to Anthropic’s documented IDE behavior.

If your only question is “how do I install Claude Code in Cursor?”, stop here. This is the answer.

What You Get Inside Cursor

Once installed, Claude Code in Cursor behaves like Claude Code in a supported IDE surface, not like Cursor’s native autocomplete engine.

The useful features are task-level features:

• multi-file edits with reviewable diffs;
• project instructions via CLAUDE.md;
• slash commands such as /init, /compact, and /clear;
• file and folder context attachment from the editor;
• MCP-backed tools when your Claude Code config enables them.

This is why the extension fits developers who think in tasks such as:

• “refactor this service layer”;
• “trace why this test is flaky”;
• “add auth checks and update the tests”;
• “explain this migration before I approve it.”

That is different from the normal Cursor rhythm of fast inline completions plus editor-native chat. Claude Code is not trying to be a tab-completion engine. It is trying to execute a coding task with more autonomy.

What the Extension Does Not Replace

This distinction matters because many developers search for this extension expecting Cursor’s native AI behavior plus Claude branding.

That is not what the Anthropic extension does.

It does not replace:

• Cursor Tab autocomplete;
• Cursor’s own model picker and agent surface;
• your normal editor workflow;
• the need to review diffs before you merge code.

The simplest way to think about it is this:

• Cursor native AI is editor-first.
• Claude Code is agent-first.

You can keep both in the same environment, but they solve different problems.

A practical split looks like this:

• Use Cursor native tools for quick edits, autocomplete, and local code exploration.
• Use Claude Code when you want a more opinionated agent loop for larger tasks.

How to Use Novita AI With Claude Code in a Cursor Workflow

This is where the setup changes.

If you want to use Novita AI as the backend, the cleanest approach is usually not “install a different Cursor extension.” It is to keep Cursor as the editor and run Claude Code from Cursor’s integrated terminal with Claude-compatible environment variables set before launch.

That gives you three useful things at once:

• Cursor stays your main IDE.
• Claude Code stays your task agent.
• Novita AI becomes the model backend.

A typical shell setup looks like this:

export ANTHROPIC_BASE_URL="https://api.novita.ai/anthropic"
export ANTHROPIC_AUTH_TOKEN="<your-novita-api-key>"
export ANTHROPIC_MODEL="<model-id>"
export ANTHROPIC_SMALL_FAST_MODEL="<model-id>"
claude

This path is the better fit if you care about model routing more than extension packaging.

It is also the safer editorial recommendation when you want to introduce Novita naturally. We are not claiming Novita replaces Cursor or ships a separate Claude Code plugin. The value is at the API and execution layer:

• one Anthropic-compatible endpoint;
• access to hosted open-source coding models in the Novita catalog;
• the option to shift heavier execution into Agent Sandbox.

Which Novita Model Path Makes Sense for Coding Tasks

If you want to add Novita to a Claude Code workflow, choose the model path based on the task instead of chasing a single “best” model.

The current Novita model catalog includes coding-relevant options such as Qwen3 Coder Next and GLM-4.7, alongside other agent-friendly models. That matters because Claude Code workflows are not all the same:

• some tasks need cheap fast iteration;
• some need longer reasoning over a large repository;
• some need better tool-use discipline over long sessions.

A practical decision rule:

• Use a lighter, cheaper coding model for short edits, code explanation, or repetitive cleanup.
• Use a stronger reasoning model for repo-wide refactors, debugging chains, or long-running task plans.
• Re-test your default model when the task pattern changes. The best model for “rename this API surface” may not be the best one for “plan a multi-step migration and update tests.”

If you want to stay close to Anthropic’s default experience, keep Claude-hosted models in the extension and reserve Novita-backed runs for workflows where cost control or open-model access actually matters.

If you want to standardize on Novita sooner, start by testing one focused coding workflow in the terminal path:

1. open Cursor;
2. open the repository terminal;
3. export the Novita environment variables;
4. run claude;
5. compare the result quality on a real task, not a toy prompt.

That gives you a clean operational answer without overcommitting.

When to Add Agent Sandbox

Model routing solves only one half of the problem. The other half is where the agent executes code.

For many developers, local execution is fine. But once Claude Code starts doing heavier work, the risk profile changes:

• dependency installs can pollute your workstation;
• long test suites block your machine;
• generated files can leave messy local state;
• parallel agent runs become hard to reproduce.

That is where Novita Agent Sandbox fits naturally. Instead of treating your laptop as the execution substrate for every agent step, you can move heavier runs into isolated environments while keeping Cursor as the day-to-day editing surface.

This setup makes the most sense when:

• you want repeatable execution for agent tasks;
• you are testing unfamiliar repositories;
• you need isolation for automation or CI-like workflows;
• your team wants separation between local editing and agent execution.

So the stack becomes:

• Cursor for editing and code review;
• Claude Code for task execution;
• Novita AI for model routing;
• Novita Agent Sandbox for isolated runtime.

That is a more durable architecture than trying to force every workflow through one IDE panel.

Recommended Workflow by Use Case

If you are deciding between paths, use this shortcut:

Your goal	Best path
Install Claude Code in Cursor quickly	Official Anthropic extension in Cursor
Keep Cursor as editor but use Novita as backend	Run Claude Code from Cursor’s integrated terminal
Try open-source coding models without changing your editor	Use the terminal path with Novita AI
Run heavier agent jobs away from your laptop	Add Novita Agent Sandbox
Keep fast autocomplete plus agentic task execution	Use Cursor native AI and Claude Code side by side

FAQ

Is Claude Code built into Cursor?

No. Cursor has its own AI features. Claude Code is a separate Anthropic product that can run inside Cursor through the extension path, or alongside Cursor through the terminal path.

Is the Claude Code Cursor extension different from the VS Code one?

Not in the way most people mean. The supported install target is the official Claude Code extension from Anthropic, and Cursor can host it because it is compatible with much of the VS Code extension ecosystem.

Should I use the extension or the terminal path?

Use the extension if you want the documented Claude Code IDE experience. Use the terminal path if backend flexibility, model routing, or Novita integration is the bigger priority.

Can I keep using Cursor’s native features after installing Claude Code?

Yes. Installing Claude Code does not remove Cursor’s native AI features. The two workflows are separate enough that many developers use both.

Is Novita AI only useful if I stop using the extension?

No. Novita is useful whenever you want control over the model and execution layer. In practice, the integrated-terminal path is simply the clearest operational way to combine Cursor, Claude Code, and Novita without pretending there is a special Novita-specific Cursor plugin.

What is the safest way to test this setup?

Start with a small real repository task. Install the extension if you want the sidebar UX. Then run a second task through Claude Code in Cursor’s terminal with Novita as the backend. Compare cost, latency, and output quality on the same kind of work you actually do.

Recommended Articles

• Claude Code vs Cursor: Which AI Coding Tool Should You Use in 2026?
• How to Use Claude Code in VS Code: LLM Backend, Model Selection, and Agent Sandbox
• Claude Code Router: How to Change Models, Providers, and Defaults

Sources checked August 10, 2026: Anthropic Claude Code IDE integrations, Anthropic Claude Code settings and environment variables, Novita AI LLM API, Novita AI model catalog, Qwen3 Coder Next on Novita AI, GLM-4.7 on Novita AI, and Novita Agent Sandbox.