Claude Code Ultraplan: Cloud-Based AI Planning in 2026 — A Hands-On Tutorial

TL;DR Summary

• Ultraplan offloads Claude Code's planning phase to a cloud session, keeping your terminal free while a structured plan is drafted remotely
• You review plans in your browser with inline comments, emoji reactions, and section-level navigation — a richer surface than terminal text
• Three ways to launch: the /ultraplan command, the ultraplan keyword in any prompt, or from a local plan's approval dialog
• You choose where to execute: in the cloud (with PR creation) or teleport back to your terminal (with full local environment access)
• Requires Claude Code v2.1.91+, a GitHub repo, and a Claude.ai account. Not available on Bedrock, Vertex, or Foundry

Direct Answer Block

Ultraplan is Anthropic's research preview feature that separates AI planning from execution by drafting structured plans in a cloud-based Claude Code session. You type a task in your CLI, Claude researches and drafts a plan remotely on Anthropic's infrastructure, and you review the plan in your browser — commenting on specific sections, asking for revisions, then choosing whether to execute in the cloud or pull the plan back to your terminal.

Introduction

Most AI coding tools conflate planning and execution. You describe a task, the agent starts editing files immediately, and if the plan is wrong you discover it 15 minutes into a broken implementation. Ultraplan breaks that cycle. It gives you a browser-based review surface where you can inspect every section of a plan, comment on specific parts, and iterate before a single line of code changes. For complex multi-step changes — migrations, refactors, architectural shifts — this changes the review dynamic entirely.

What is Anthropic Ultraplan and what problem does it actually solve?

Ultraplan is not just "plan mode in the cloud." It's a structural separation of planning from execution that solves a specific terminal UX limitation: long plans are hard to review in a 24-line scrollable prompt window.

"You write a task in the CLI, and Claude drafts a structured plan in a cloud session you can review in your browser. This separates planning from execution and gives you a clearer way to inspect and edit multi-step changes before running them." — AlphaSignal summary of Anthropic's Ultraplan announcement

The core problem it addresses is plan review friction. In local plan mode, Claude produces a plan in the terminal — you read paragraphs of text, type a response, and Claude re-drafts. If you want to comment on a specific section, you have to quote it or describe its position. With Ultraplan, the plan appears in a web interface where you can:

• Highlight any passage and leave an inline comment for Claude to address
• React with emojis (thumbs up, thinking face) to signal approval or concern without writing full feedback
• Jump between sections via an outline sidebar

This matters most for changes spanning 5+ files with architectural implications. The terminal review of such a plan takes patience; the browser review takes seconds.

According to Anthropic's docs, the cloud session runs on your account's default cloud environment. If you don't have one, Ultraplan creates it automatically on first launch.

How do you launch Ultraplan from the CLI (and what are the three ways to trigger it)?

There are three trigger methods, from explicit to incidental:

Method 1: The /ultraplan command (explicit)

/ultraplan migrate the auth service from sessions to JWTs

This is the most intentional path. Type the slash command with your task, confirm the dialog, and Claude launches a remote session. The CLI shows a status indicator:

Method 2: The ultraplan keyword (implicit)

Include the word "ultraplan" anywhere in a normal prompt:

Help me plan a refactor of the payment service — use ultraplan

Same result, less typing. This path also shows a confirmation dialog before launching.

Method 3: From a local plan (iterative)

When Claude finishes a local plan and presents the approval dialog, select No, refine with Ultraplan on Claude Code on the web. This sends your existing draft to the cloud for richer iteration. This path skips the confirmation dialog since selecting the option is already consent.

Run /tasks at any point to see the Ultraplan entry, open detail view with the session link, or stop the plan. Stopping archives the cloud session and clears the indicator; nothing is saved to your terminal.

Important constraint: If Remote Control is active, it disconnects when Ultraplan starts because both features occupy the claude.ai/code interface — only one can be connected at a time.

How does reviewing and iterating on a plan work in the browser?

When the status changes to ◆ ultraplan ready, open the session link. The plan appears in three zones:

1. The plan document — a structured breakdown of the proposed changes, organized by sections (migration steps, file changes, testing strategy, risks)
2. Inline comments — highlight any text, leave a comment, and Claude revises that specific section in response
3. Outline sidebar — navigable section index for jumping between parts of the plan

Here's the iteration loop:

• You highlight a section like "Proposed database migration: 3-step rollout with rollback" and comment: This assumes zero-downtime. Can we add a step for staging validation first?
• Claude revises the plan, inserting the staging validation step
• You react with a thinking face emoji on the rollback strategy, signaling uncertainty
• Claude proposes an alternative rollback mechanism
• You approve

This is fundamentally different from terminal-based iteration. In the terminal, you'd need to: read the full plan, type a revision request covering multiple sections, hope Claude understood which sections you meant. In the browser, your feedback is surgically attached to specific text.

The plan document also supports emoji reactions at the section level. These are lightweight signals — a thumbs up means "this section looks right," a thinking face means "reconsider this" — that let you communicate without typing.

Should you execute your Ultraplan in the cloud or teleport it back to your terminal?

When the plan is approved, you pick from two execution paths from the browser:

Execute on the web

Select Approve Claude's plan and start coding to have Claude implement the plan in the same cloud session. Your terminal shows a confirmation, the status indicator clears, and work continues in the cloud. When done, you review the diff and create a pull request from the web interface.

Best for: When you don't need local access (environment variables, private dependencies, local services). The cloud session has your repo but not your machine's runtime.

Teleport back to terminal

Select Approve plan and teleport back to terminal to pull the plan into your local CLI session. The cloud session archives, and your terminal shows three options:

• Implement here: inject the plan into your current conversation
• Start new session: clear context, begin fresh with only the plan
• Cancel: save the plan to a file without executing (Claude prints the file path)

If you start a new session, Claude prints a claude --resume command so you can return to your previous conversation.

Best for: When you need local environment access, private dependencies, or are running integration tests against local services. The plan lands in your terminal with full access to your machine.

When should you use Ultraplan instead of local plan mode (and when should you NOT)?

Use Ultraplan when:

• The change spans 5+ files with architectural implications — you need a rich review surface
• You want hands-off drafting — Ultraplan runs remotely, your terminal stays free for other work
• You're on a team and want async plan review — share the browser link, get comments before execution
• The plan needs multiple iterations — inline comments are faster than terminal-based revision prompts

Use local plan mode (/plan or Shift+Tab into plan mode) when:

• The change is small and self-contained — 2-3 files, quick to review in terminal
• You're on Bedrock, Vertex, or Foundry — Ultraplan requires Anthropic direct API and is not available on these providers
• You don't have a Claude.ai web account — Ultraplan runs on Claude Code on the web infrastructure
• You want instant iteration — local plan mode has no remote session startup time (typically 15-30 seconds for Ultraplan)

Do NOT use Ultraplan when:

• Your organization requires Zero Data Retention (ZDR) — Ultraplan runs on cloud infrastructure where ZDR is not available
• Your repository is not on GitHub — the cloud session needs a GitHub remote to clone and operate
• You're working on sensitive code that cannot leave your machine — the repo is bundled and uploaded to Anthropic's cloud sandbox

How does Ultraplan compare to Ultrareview, and when should you use both?

Ultraplan and Ultrareview are siblings: one plans before work, the other reviews before merge.

The ideal workflow pairs both:

1. /ultraplan — plan a complex feature in the browser, iterate on architecture
2. Implement — execute in cloud or terminal
3. /ultrareview — run a multi-agent deep review before merging

Ultrareview has one distinct advantage over local /review: every reported finding is independently reproduced and verified by the agent fleet, so results focus on real bugs rather than style suggestions. It supports both branch diff mode (reviews changes against default branch) and PR mode (clones the PR from GitHub directly).

Ultrareview includes a non-interactive mode for CI: claude ultrareview runs headless, prints findings to stdout, and exits with code 0 on success or 1 on failure. Pass --json for raw output or --timeout <minutes> to limit wait time.

Frequently Asked Questions

Q: Do I need a paid Claude subscription to use Ultraplan?

Ultraplan requires a Claude Code on the web account, which is tied to a Claude subscription (Pro, Max, Team, or Enterprise). It's not available with API-key-only authentication.

Q: What happens to my code during Ultraplan?

Your repository state is bundled and uploaded to Anthropic's cloud sandbox for plan drafting. The sandbox is ephemeral — destroyed when the session ends. For Ultrareview in PR mode, the sandbox clones directly from GitHub rather than uploading your local state.

Q: Can I use Ultraplan without a GitHub repo?

No. The cloud session needs a GitHub remote to clone and operate on your codebase. If your repo is on GitLab or Bitbucket, Ultraplan is not available.

Q: How much does Ultraplan cost?

Ultraplan counts toward your plan's included usage. It does not bill as extra usage like Ultrareview does. You pay only in token consumption from the planning session.

Q: Can multiple people review the same Ultraplan?

Yes. Share the browser session link with teammates. They can view the plan, leave comments, and react to sections. Only one person's feedback drives Claude's revisions at a time, but multiple people can participate in the review.

Q: Is Ultraplan available on the VS Code extension?

Ultraplan is launched from the CLI. If you're using Claude Code inside VS Code's integrated terminal, the /ultraplan command works there too. The browser review interface is separate from VS Code.

Glossary

• Ultraplan: A cloud-based planning feature that drafts Claude Code plans remotely, reviewable in a browser
• Ultrareview: A cloud-based code review feature that uses multiple AI agents to find and verify bugs before merge
• Plan mode: Local Claude Code mode (/plan or Shift+Tab) that researches and proposes changes without editing files
• Teleport: The mechanism for pulling a cloud-drafted plan back into a local terminal session for execution
• Cloud session: A Claude Code session running on Anthropic's managed infrastructure, not your local machine
• Ultraplan ready: The status indicator confirming the cloud-drafted plan is available for browser review

Author

Ramsis Hammadi — AI/ML engineer specializing in GenAI, LLM engineering, and automation. Full bio →