Cursor CLI in NeuroSquadCursor’s agent,many at once, on one canvas
Every card runs the real cursor-agent in its own terminal, signed in with your Cursor account. NeuroSquad knows when each one is working, waiting for your approval or done, resumes its chat after a restart, and connects it to terminals, browsers, notes and other agents by arrow.
Official sitecursor.com/cli- The real Cursor CLI, your Cursor login
- Nothing written to
~/.cursor - Free, for Windows 10 and 11
Working, waiting on you, done — without guessing from silence
Cursor runs a plugin’s hooks for tools, but not for the start or end of a turn. So NeuroSquad reads three signals Cursor itself produces: its running indicator, its preToolUse hook and the turn_ended line of its own transcript.
ctrl+c to stopsignalYou press Enter and Cursor’s prompt bar shows “ctrl+c to stop” (or its transcript grows). The card’s icon spins and the Squad status card counts it as working.
preToolUsesignalThe plugin’s preToolUse hook names the tool — “Shell — rm -rf dist” — and Cursor’s own approval dialog on screen confirms it is really asking. The card pulses and the toast says what it wants.
turn_endedsignalCursor writes {"type":"turn_ended"} into the chat’s transcript — success, error or aborted. The next queued prompt goes out and a toast says it’s done.
A real end, not a pause
The turn ends when Cursor records it, so a long think never passes for “done”. A request that fails before writing anything is caught by the indicator disappearing.
Cancelled turns too
Ctrl+C makes Cursor write turn_ended with aborted: the card finishes, the process stays alive.
Click, and you’re there
A toast click switches to the workspace, flies the camera to the card and flashes it — even among dozens of agents.
Three things you’ll do every day
Replicas of the app, played step by step. When Cursor asks for approval, the answer is yours.
Cursor fixes a failing test, stops for approval before deleting a folder, and finishes — the journal writes itself.
What Cursor CLI can do in NeuroSquad — and what it can’t
Most of what NeuroSquad does for Claude Code works for Cursor too. Where Cursor has no way to do it, the card below says so instead of pretending.
- 01
Knows its state
Built from Cursor’s own transcript, hooks and screen, not from a quiet terminal.
- Emulated
Turns read from Cursor itself
Cursor runs no plugin hook when a turn starts or ends, so the start comes from its “ctrl+c to stop” indicator and the end from the
Docs: Turns read from Cursor itselfturn_endedline it writes to its transcript. A notification that names the tool
The
Docs: A notification that names the toolpreToolUsehook names what waits for approval —Shell — rm -rf dist,MCP — note_write— and the toast leads with it. One per agent; a click flies to the card.Every card is its own chat
The first launch passes Cursor’s
Docs: Every card is its own chat--new-session-idwith the card’s id, later ones--resume, so a restart continues the chat./clearand/resumeare followed through thesessionStarthook.The Squad status card
Every agent of the workspace with its status and live turn time. The model is the chat’s last one from Cursor’s own store, or the default from your
Docs: The Squad status cardcli-config.json.Custom cards can follow it
Cards built with the Card SDK see the same statuses and turn events, and read its last reply from Cursor’s transcript.
Docs: Custom cards can follow it
- 02
Works through the canvas
An arrow from the card is a set of tools, served by NeuroSquad’s MCP server through a per-card Cursor plugin.
- Emulated
Arrows are tools
Cursor lists a server’s tools once, when it connects, so it is offered NeuroSquad’s built-in tools from the start and each arrow decides at call time what works. Tools of MCP or skill cards connected later arrive on the next start.
Docs: Arrows are tools It can add cards itself
Docs: It can add cards itselfcanvas_spawn_cardlets it open a terminal, a browser, a note or a helper agent — another Cursor or any other CLI — next to itself, already joined by an arrow.Canvas mode
NeuroSquad answers Cursor’s
Docs: Canvas modepreToolUsefor Shell, WebFetch and WebSearch with a refusal, so commands run in terminal cards you can watch. It holds even under--yolo, and the switch works on a running card.MCP servers and skills
Servers from the official MCP Registry and skills from skills.sh reach Cursor through the card’s plugin. Cursor asks before every call to an installed server; their secrets never leave NeuroSquad.
Docs: MCP servers and skills- Not possible
Tool calls stop at 60 seconds
Cursor gives up on any MCP call after 60 seconds, and its config has no way to raise that. NeuroSquad tells it to wait on other agents in slices of 50 seconds and call again.
Docs: Tool calls stop at 60 seconds
- 03
Built for long runs
Context, limits, queued prompts and hand-offs, handled on the card.
Context meter and Compact
The ring in the header reads the chat’s own token count from Cursor’s store — the same percentage Cursor shows. Compact sends
Docs: Context meter and Compact/summarize.Usage limits flagged
When Cursor’s backend refuses a request for a usage limit, the card shows it. Automatic “continue” needs a reset time, and Cursor’s current message doesn’t give one.
Docs: Usage limits flaggedPrompt queue
Line up the next messages: each goes out when the turn has ended. After a cancelled turn Cursor puts your prompt back in its box — a queued one is added after it.
Docs: Prompt queueHand off to a fresh agent
When a chat is full, start a new agent beside it — Cursor or any other CLI — with a summary and Cursor’s last answer typed in, ready for you to send.
Docs: Hand off to a fresh agentJournal
Each finished answer, read from Cursor’s transcript, is added to a connected note: a running log of what the agent did.
Docs: Journal
- 04
Works in a team
Several agents — Cursor or a mix of CLIs — on one project without stepping on each other.
Isolated worktrees
Give an agent its own git worktree, so parallel agents never edit the same files. A setup command and files to copy, like
Docs: Isolated worktrees.env, prepare it first.Add a reviewer
One click starts an agent on a different CLI in the same worktree, with an arrow from the author and a prompt to review without editing anything.
Docs: Add a reviewerDangerous mode, per card
For a run you trust, a card starts Cursor with
Docs: Dangerous mode, per card--yolo— Cursor’s “Run Everything”. Off by default; canvas mode still holds.Roles
Reviewer, tester, researcher, architect, writer or your own: the role is typed in as a first prompt you can see and edit.
Docs: RolesOne instructions file
Cursor reads AGENTS.md and CLAUDE.md itself, so the file you already keep for your other agents works for Cursor as is.
Docs: One instructions file
- 05
Cost, control and access
Here Cursor differs most: its models and billing live on Cursor’s servers.
- Not possible
No usage or cost tracking
Cursor keeps no per-request token counts on disk — only in a hook plugins can’t receive — and bills on its own servers. Its cards don’t appear in NeuroSquad’s Usage section.
Docs: No usage or cost tracking - Not possible
No OpenRouter models
The Cursor CLI talks only to Cursor’s backend: its custom-endpoint flags belong to a separate build, and
Docs: No OpenRouter models--modeltakes only your account’s models. A Cursor card runs on your Cursor plan. Budget brake still stops it
Cursor’s own spend isn’t counted (see above), but when the workspace budget runs out it gets Ctrl+C with every other agent and automatic prompts stop.
Docs: Budget brake still stops itDictation into the prompt
Hold a hotkey and speak. Speech is recognised on your machine and typed into the active agent’s prompt.
Docs: Dictation into the promptFrom your phone
Scan a QR code and the whole canvas opens in the phone’s browser: read Cursor’s terminal, answer its approval, send the next prompt.
Docs: From your phone
Who’s waiting — and the honest part about cost
The Squad status card works for Cursor as for any agent. The Usage section doesn’t, and here is why.
The Squad status card
Every agent of the workspace with its status, model and turn time. The one waiting on you comes first, with the tool it wants to run. Tap a status to filter.
- coupon-fixCursor CLIcomposer-2.5Shell — rm -rf distNeeds you1:17
- cart-testsCursor CLIcomposer-2.5Cover the cart with testsWorking9:33
- search-uiCursor CLIgpt-5.5Rework the search boxWorking3:51
- docs-passCursor CLIcomposer-2.5Update the READMEFinished6 min. ago
Usage & cost
NeuroSquad counts usage from each CLI’s own logs on your disk. Cursor writes none that carry tokens.
The real Cursor CLI, plus a plugin and a few arguments
NeuroSquad starts the Cursor CLI you installed — its newest version’s own node.exe, directly — in a real terminal in the card. This is everything it adds:
Nothing written to ~/.cursor
The plugin and the config overlay are per-card folders in NeuroSquad’s data. Your chats stay in your own chat store.
No secret on disk
The MCP token lives only in the card’s environment; Cursor expands it itself.
Hooks that actually run
Cursor runs hooks through Windows PowerShell and a Git Bash environment breaks them — the card clears those variables, and NeuroSquad reads the hooks’ answers despite PowerShell’s BOM.
Cursor CLI in NeuroSquad, answered
Give Cursor’s agent a canvas
Free for Windows 10 and 11. Your code and your Cursor login stay on your machine.