How to Use Claude Code Ultra Plan: Setup and Workflow
Ultraplan is Claude Code's cloud planning feature. Connect GitHub with /web-setup, trigger with /ultraplan, review the plan on the web, then build.
What Ultraplan actually is
Claude Code is Anthropic’s terminal coding agent. It runs in your terminal, reading and writing files, running shell commands, and working through multi-step development tasks on its own.
Ultraplan is a feature inside Claude Code, not a subscription tier or a pricing plan. It’s a cloud-based, multi-agent planning mode you trigger for a specific task. Instead of planning locally in a single pass, Claude Code sends multiple sub-agents into the cloud to explore your codebase at the same time, then gives you a structured, multi-step plan in a web interface. You review, comment on, and edit that plan before any code is written, then execute it in the cloud or back in your terminal.
This guide walks through the setup, the exact commands, and the full workflow from trigger to execution.
Before you start
Ultraplan has a few prerequisites. Get these in place first.
Install and update Claude Code
Claude Code ships as an npm package, so you need Node.js (v18 or later) and terminal access. Install it globally:
npm install -g @anthropic-ai/claude-codeThat gives you the claude command. Ultraplan needs Claude Code v2.1.91 or later, so check your version and update if you’re behind:
claude --versionOn first run, claude opens a browser window to sign you into your Claude account.
Connect your GitHub repo
Ultraplan’s sub-agents run in the cloud, so they need access to your code through GitHub. Inside Claude Code, run:
/web-setupRemy is new. The platform isn't.
Remy is the latest expression of years of platform work. Not a hastily wrapped LLM.
This links your GitHub repository so the cloud agents can read it when they plan.
Commit and push your work
The cloud agents plan from your GitHub remote, not from uncommitted files on your local disk. Commit and push your latest changes before you trigger a plan — otherwise the sub-agents reason about stale code and the plan won’t match what you’re actually working on.
git add -A && git commit -m "wip" && git pushHow to trigger Ultraplan
There are three ways to start an Ultraplan session:
- Run the command directly. Type
/ultraplan <task>with a plain-language description — for example,/ultraplan migrate the auth service to JWTs. - Mention it in a prompt. Type “ultraplan” anywhere in a normal prompt and Claude Code will route the task through the cloud planner.
- Refine an existing plan. If a quick local plan comes back too shallow for the task, ask Claude to “refine with Ultraplan” to send it to the cloud for deeper exploration.
Once triggered, Claude Code hands the task to multiple sub-agents. They explore different parts of your codebase in parallel — one tracing the database layer, another the API routes, another the frontend — and report back so the plan covers cross-file dependencies and edge cases a single pass would miss.
Reviewing and editing the plan
Ultraplan produces its plan in a web interface rather than a scroll of terminal output. Open the session link at claude.ai/code to see it.
The plan is structured into steps you can work through:
- Read the full plan before approving anything. This is the cheapest point to catch a wrong assumption.
- Comment on specific sections where something looks off or needs clarification.
- Edit and tweak the steps directly — reorder, cut, or adjust them until the plan matches what you actually want built.
Nothing gets written to your codebase during this stage. You’re shaping the plan while it’s still just a plan.
Executing the plan: cloud or local
Once the plan looks right, you approve it and choose how to build:
- Approve plan and start coding on the web. The work runs in the cloud, executing the approved steps against your connected repo.
- Teleport back to the terminal. Send the approved plan to your local Claude Code session and execute it there, against your local environment.
Either way, Claude Code is now working from a plan you’ve reviewed and signed off on, which is what makes a large refactor far less risky than letting an agent improvise across dozens of files.
When to use Ultraplan (and when to skip it)
Ultraplan adds coordination overhead, so it’s not the right tool for every task.
When it’s worth it
- Large codebases with complex interdependencies, where a change in one file ripples through many others.
- Cross-cutting changes like dependency upgrades, auth refactors, API versioning, or design-system migrations.
- Open-ended or uncertain work, where you want the planner to weigh several approaches before committing.
- High-stakes production code, where catching a bad assumption at the plan stage saves a painful rollback later.
When to skip it
- Small, isolated tasks — fixing a bug, writing one function, updating docs.
- Greenfield projects with no existing complexity to navigate.
- Work you’ve already analyzed, where you just need execution, not exploration.
- Rapid prototyping, where iteration speed matters more than plan quality.
Everyone else built a construction worker.
We built the contractor.
One file at a time.
UI, API, database, deploy.
For those cases, a quick local plan is faster because it skips the cloud round trip.
Getting better plans
A few habits make Ultraplan’s output stronger:
Scope your request. “Refactor the auth module to use JWT instead of sessions” plans far better than “refactor the whole codebase.” Clear boundaries give the sub-agents something concrete to explore.
Keep a CLAUDE.md file. Claude Code reads a CLAUDE.md in your project root for conventions, architecture notes, and things it shouldn’t touch. The more you tell it there, the less the sub-agents have to infer.
Push before you plan. Since the plan is built from your GitHub remote, make sure your latest work is committed and pushed so the sub-agents aren’t planning against stale code.
Use the review stage. Don’t rubber-stamp the plan. Read it, comment where something’s off, and edit before approving — the web interface exists so you can fix problems before they become code.
Extending Claude Code with MindStudio
Claude Code handles writing and reasoning about code well. On its own, though, it can’t send emails, query external APIs, generate images, trigger automations, or talk to most business tools.
That’s where the MindStudio Agent Skills Plugin helps. It’s an npm SDK (@mindstudio-ai/agent) that gives Claude Code agents access to 120+ typed capabilities as simple method calls, without building integrations from scratch.
Instead of Claude Code reasoning through how to wire up an email integration or a Slack notifier, you drop the SDK in and call methods like agent.sendEmail(), agent.searchGoogle(), or agent.runWorkflow(). The SDK handles the infrastructure layer — rate limiting, retries, authentication, error handling — so the agent focuses on logic, not plumbing.
For workflows built around Ultraplan sessions, where the work spans code plus other systems, this keeps Claude Code from having to reinvent every integration itself.
MindStudio itself is a no-code platform for building AI agents and automated workflows, but for Claude Code users the Agent Skills Plugin is the most relevant piece. It’s an npm package and works with any agentic system, MindStudio-built or not.
You can try MindStudio free at mindstudio.ai.
Frequently asked questions
Is Ultraplan a paid tier?
No. Ultraplan is a planning feature inside Claude Code, not a subscription level. You trigger it per task with /ultraplan; it’s not a plan you buy that unlocks rate limits.
Do I need a GitHub repo to use Ultraplan?
Yes. The cloud sub-agents plan from your GitHub remote, so you connect your repository with /web-setup and push your latest commits before triggering a plan. Planning against uncommitted local files won’t work.
What version of Claude Code do I need?
Claude Code v2.1.91 or later. Check with claude --version and update via npm install -g @anthropic-ai/claude-code if you’re behind.
How do I actually start a plan?
Run /ultraplan <task> in Claude Code — for example, /ultraplan migrate the auth service to JWTs. You can also type “ultraplan” in a prompt, or ask Claude to “refine with Ultraplan” when a local plan is too shallow.
Where do I review the plan?
One coffee. One working app.
You bring the idea. Remy manages the project.
In the web interface at claude.ai/code. Open the session link, read the structured plan, comment on sections, and edit the steps before you approve.
Can I run the plan on my own machine instead of the cloud?
Yes. After approving, you can either start coding on the web (running in the cloud) or teleport the approved plan back to your local terminal and execute it there against your local environment.
Is Claude Code safe to run on a production codebase?
Use caution. Claude Code can read, write, and delete files and run shell commands. Work on a feature branch, review the plan carefully before approving, and review all diffs before merging.
Key takeaways
- Ultraplan is a cloud-based, multi-agent planning feature in Claude Code — not a subscription tier.
- Setup: Claude Code v2.1.91+, connect GitHub with
/web-setup, and commit and push your work so the cloud agents plan from your remote. - Trigger with
/ultraplan <task>, by mentioning “ultraplan” in a prompt, or by asking Claude to “refine with Ultraplan.” - Review and edit the structured plan in the web interface at claude.ai/code before approving anything.
- Execute the approved plan in the cloud, or teleport it back to your terminal to run locally.
For extending what your Claude Code agents can do beyond writing code, MindStudio is worth a look.
