How to run Cursor and Claude Code together
You can install Anthropic’s Claude Code extension in Cursor or run the standalone Claude Code CLI from its terminal. Claude Code is not the same as Cursor Agent using a Claude model, and running both does not create automatic coordination. BAND (band.ai) for desktop lists both Cursor and Claude Code integrations for shared-room handoffs.
Reviewed November 2026
Which two products are you actually running?
Cursor is an editor with its own agent surfaces; Claude Code is Anthropic’s coding agent with terminal and IDE interfaces. A Claude model selected inside Cursor Agent does not turn that session into Claude Code. The distinction matters for tools, conversation state, permissions and billing. Name the actual agent when assigning work so the reviewer knows which execution path produced the change.
Anthropic’s VS Code documentation includes an Install for Cursor option. That establishes a supported extension installation route, not a shared subscription or automatic communication between the two agents. Keep each session’s purpose visible: one writes, one investigates, or both work on separate tasks with a planned integration step.
How do you install and sign in to Claude Code?
Use the Install for Cursor path in Anthropic’s extension documentation. The extension provides a graphical Claude Code interface with inline diffs, plan review, file mentions, history and multiple conversations. Follow the documented authentication flow for a qualifying Claude subscription or Claude Console account; supported third-party-provider configurations have their own instructions.
The extension bundles a CLI for its chat panel. Running the claude command in an integrated terminal additionally requires the standalone CLI installation. Do not assume an extension install has also prepared every terminal on the machine. Verify the intended interface in the repository before starting a second agent on a related task.
Which execution layout fits your task?
| Layout | Use it for | Coordination | Main risk |
|---|---|---|---|
| Claude Code extension in Cursor | One Claude Code task in the editor | Developer directs the session | Confusing it with Cursor Agent |
| Two agents, one checkout | Read-only investigation or sequential edits | Developer controls ownership | Overlapping writes |
| Separate worktrees | separate implementation tasks | Explicit integration handoff | Conflicts when combining branches |
| BAND rooms | Dependent work across listed agents | @mention handoffs and shared context | Permissions and review still matter |
Choose the layout before prompting. For a read-only review, one checkout can be sufficient if the reviewer does not modify files. For two implementations, isolate their changes. If the second task cannot begin until the first defines an interface, make that dependency explicit rather than trying to hide it with more concurrency.
How do you stop the agents editing the same files?
Assign one owner per changing area. A writer owns the implementation; a reviewer reports findings and asks for revisions. If you want separate implementations, use separate branches and worktrees. Claude Code documents --worktree for named isolated sessions; Cursor documents its worktree workflow in the Agents Window and Worktree Skills commands for the IDE.
Worktrees share repository history but have separate working directories. They do not automatically isolate databases, ports or remote services. Give each environment its own configuration where concurrent tests could interfere. Keep credentials out of task messages and generated example files, and review permissions before an agent can push or alter shared infrastructure.
What should the handoff between Cursor and Claude contain?
Include the task goal, exact branch and revision, changed files, the intended interface and the test evidence. A statement such as “the backend is done” is not enough for another agent to build a compatible frontend. Tell the receiver what it should verify and what remains a human decision. Mark any tests that were not run instead of presenting an unverified change as complete.
For example, Cursor can prepare an API implementation and Claude Code can review error handling against the accepted contract. That is a workflow example, not a product benchmark. The review agent needs the candidate revision, not merely the writer’s summary. If it finds a missing case, return the finding to the writer with a concrete acceptance check.
What changes when both agents join a BAND room?
BAND lists Cursor and Claude Code as coding-agent integrations. Local and centrally managed agents share rooms, memories and an activity timeline, with messages routed by @mention. People approve, redirect or override in place. This adds a place for the handoff and its discussion, rather than making the developer carry every message between separate sessions.
Follow the current integration setup for each agent and verify a small handoff before using it for a large change. A room is a coordination layer, not proof that both agents have access to the same branch or tools. Keep a human-owned merge gate, especially for migrations, permission changes and dependency updates.
How do subscriptions and costs stay separate?
Installing Claude Code inside Cursor does not establish that Cursor pays for Anthropic usage. Use the official authentication and pricing documentation for each product. This guide does not publish a combined token allowance or an unverified current Cursor plan price. Multiple simultaneous sessions can consume usage even when the human sees them in one editor.
BAND Free is $0/month for a single user’s personal non-commercial use, with two-week retention. Pro is $17.99/month for single-user personal use with on-demand chat export. Multiple users and teams need Enterprise by quote with custom retention. Those terms describe coordination; they do not replace the execution costs or subscriptions of the coding agents.
Where should you go next?
Read the Cursor review, Claude Code agent teams review and coding agent handoff guide before adding more sessions. The Claude Code and Codex guide shows the same coordination problem with another vendor.
Start with one writer and one read-only reviewer. Check whether the second agent finds useful issues and whether the human can follow the change history. Add parallel implementation only when the task boundaries and integration procedure are clear enough that extra execution does not create extra ambiguity.
What else do readers ask?
Can Claude Code run inside Cursor?
Yes. Anthropic’s extension documentation includes an Install for Cursor path.
Is Cursor Agent using Claude the same as Claude Code?
No. A model choice inside Cursor Agent is different from Anthropic’s Claude Code product.
Does installing the extension install the terminal CLI?
The extension bundles its chat-panel CLI; using claude in a terminal requires the standalone CLI installation.
Do both agents automatically share conversation history?
Do not assume shared history or automatic coordination from using the same editor.
Can both join BAND rooms?
Yes. Cursor and Claude Code are listed BAND coding-agent integrations.