Claude Code Ultraplan offloads your most complex planning tasks to a cloud Opus 4.6 session for up to 30 minutes — while you keep working locally. Here’s how to use it in five steps.

Prerequisite: Ultraplan requires the $400/month Claude Code plan tier.

Step 1: Open Ultraplan from the Command Palette

In Claude Code, open the command palette (Cmd/Ctrl + Shift + P) and type Ultraplan. Select “Claude Code: Start Ultraplan Session”.

Alternatively, from the Claude Code sidebar, look for the Ultraplan tab beneath your current session. This tab is only visible on eligible plan tiers.

Step 2: Write Your Planning Prompt

Ultraplan works best with problems that are too complex for a quick interactive answer. The ideal Ultraplan prompt:

  • Describes the scope of what you’re planning (a system, a refactor, a migration, a feature)
  • Includes constraints the plan must respect (language, dependencies, team size, timeline)
  • Specifies the desired output format (phases? architecture diagrams? risk analysis? all three?)

Example prompt:

Plan a migration of our monolith payments module to a standalone microservice.
Constraints:
- Zero downtime required
- PostgreSQL + Redis stack
- 3-engineer team, 6-week runway
- Must maintain backwards compatibility with existing API consumers during migration

Output: phased implementation plan with dependency diagram, risk flags for each phase, and a rollback strategy.

The more specific your constraints, the more actionable the Ultraplan output. Vague prompts produce vague plans.

Step 3: Submit and Keep Working

Once you submit the Ultraplan prompt, Claude Code hands the job to a Cloud Container Runtime (CCR) running Opus 4.6 in plan mode. Your local session is immediately freed.

You’ll see a status indicator in the Ultraplan tab showing the job is running. You can:

  • Continue coding in your local Claude Code session
  • Switch to a different project or workspace
  • Close your laptop (the cloud job continues)

Ultraplan jobs typically complete in 5–15 minutes for well-scoped problems. The 30-minute ceiling exists for genuinely complex planning work.

Step 4: Review the Plan

When the Ultraplan job completes, Claude Code notifies you via the sidebar status indicator (and optionally via a system notification if you’ve enabled those in Claude Code preferences).

The output document appears in the Ultraplan tab. It includes:

  • Executive summary: The core approach in 3–5 sentences
  • Architecture diagrams: Rendered Mermaid diagrams for system structure and data flow
  • Implementation phases: Ordered steps with dependencies and complexity estimates
  • Risk flags: The riskiest decisions in the plan, with alternative approaches
  • Execution plan: A machine-readable structured breakdown that Claude Code can use directly

Read the plan carefully. This is the moment to catch assumptions you disagree with, add constraints you forgot to mention, or redirect the approach before any code is touched.

Step 5: Approve or Execute

After reviewing, you have two options:

Option A — Manual execution: Use the plan as a reference document. Copy sections into your Claude Code chat as needed, paste architecture decisions into your documentation, or use the phased breakdown to assign work to your team. The plan is a rich Markdown document; treat it like any other planning artifact.

Option B — Agentic execution: Click “Execute Plan” in the Ultraplan tab. Claude Code will use the Ultraplan document as the authoritative spec for an agentic run — spawning subagents, writing code, creating files, and running tests according to the plan’s phases. You’ll be asked to approve the execution scope before anything runs.

Option B is powerful but should be used with care. Review the plan thoroughly before approving execution. If a phase has a risk flag you don’t have a mitigation for, address it manually before handing the plan back to Claude Code for execution.

Tips for Better Ultraplan Results

Be specific about constraints. Ultraplan’s Opus 4.6 instance can’t ask clarifying questions mid-run. All the constraints need to be in the initial prompt.

Include your existing architecture context. Paste in a brief description of your current system, key dependencies, and any relevant architectural decisions. Ultraplan doesn’t have access to your codebase — you need to provide that context explicitly.

Use it for architecture, not implementation details. Ultraplan shines on system-level planning. For implementation-level tasks (write this function, refactor this module), regular Claude Code interactive sessions are faster.

Save the plan document. The Ultraplan output is valuable beyond the immediate task. Export it to Markdown and commit it alongside your codebase as a decision record.

Related: Claude Code Ultraplan announcement and overview · Official Ultraplan docs