Three ways to put your AI on the plan.
Connect your local agent live over MCP. Dispatch a task to a cloud sandbox and come back to a PR. Or hand off the old-fashioned way with a one-click prompt or a full project export. However your AI works, the plan goes with it.
MCP, the live connection
Your local agent connects once and works directly against the plan. It reads scope and context, then writes status, memories, docs, and comments back as it builds.
RecommendedCloud Run, fully managed
Dispatch a task to a cloud sandbox. Claude Code or Codex picks it up, builds it, and opens a PR. No local agent running, no local setup.
BetaManual, copy and paste
No MCP? Copy a single task as a prompt and paste it into any agent, or export the entire project as JSON, Markdown, or rules.
Your AI reads the plan, and writes back to it.
One connection turns the plan into a live workspace your agent works inside. It pulls a full knowledge briefing on every session, then keeps the plan current as it goes.
Connect Claude Code, Cursor, or any MCP-capable agent. Or paste the setup to your agent and it configures itself.
Loads the full blueprint and knowledge briefing on connect, lists your projects, and updates project status as work moves.
Reads the structure, and adds or renames modules and features when it discovers work that belongs in the plan.
Pulls the next agent-ready task, moves tasks through the workflow, adds new ones, and ticks off acceptance criteria as it goes.
Reads and searches your project docs, and writes new ones back, so decisions and notes grow instead of leaking away.
Saves durable decisions and preferences, searches prior ones, and promotes the important stuff into the always-on context.
Reads the always-injected context briefing, reads and posts comments, and @mentions you when it hits a decision.
Dispatch a task. Come back to a PR.
Pick a task from your plan, dispatch it to a cloud sandbox, and walk away. Claude Code or Codex runs inside a dedicated VM, builds the task, and opens a pull request when it's done. No local agent, no terminal, no babysitting.
Pick a task
Any atomic task in your plan can be dispatched. The scope, acceptance criteria, and project context go with it automatically.
Agent builds in the cloud
Claude Code or Codex runs in a dedicated VM sandbox. It reads your repo, executes the task, and writes status back to the Command Center as it works.
PR opened, task closed
When the build finishes, the sandbox opens a pull request against your repo and marks the task complete. You review and merge.
Or hand it off in one click.
No MCP, no cloud setup. The plan still comes with you. Copy a task prompt or export the whole project, both in one click.
Copy one task as a prompt.
Every task has a Copy Prompt button. It packs the task, its acceptance criteria, and context into a clean prompt you paste into any agent.
Export the entire scope.
Hand off the whole project at once. Export as JSON, Markdown, or agent rules, tuned for Cursor, Claude Code, Devin, or generic.
Clean specs your agent can actually execute.
Hand off is a step in the loop, not the end of it. Export structured JSON or Markdown, connect MCP for a live link to Cursor and Claude Code, or use Send to Agent on a single task. Your workspace stays the system of record while the build runs.
Atomic = sized for one agent session, with explicit acceptance criteria. Roughly 30 minutes to 2 hours of work.
{
"project": "Stripe billing for SaaS",
"agent_target": "claude_code",
"tech_stack": ["Next.js", "Prisma", "Postgres"],
"modules": [
{
"name": "Billing core",
"features": [
{
"name": "Subscription lifecycle",
"tasks": [
{
"id": "task_01",
"title": "Create subscription on checkout success",
"description": "Listen for checkout.session.completed webhook, upsert subscription row, attach customer.",
"acceptance_criteria": [
"Webhook signature verified",
"Idempotent on retry",
"Subscription row created with status=active"
],
"complexity": "M",
"depends_on": []
}
]
}
]
}
]
}Hand off, and stay in the loop.
MCP for a live connection, cloud runs for a fully managed build, or copy in one click. Every path keeps the plan as the system of record.