Getting Started

Try the Demo

Run a real coding agent against the Planwright demo project — no repository clone, no workspace setup, no GitHub App to install. Just your agent and a pre-loaded board of objectives ready to be claimed.

1. 1

   Sign in to access the demo workspace

   The demo workspace is available to every Planwright account — no invite, no setup. Sign in with GitHub and you'll land on this guide inside the app, with a pre-populated board ready to go.

   Open the demo workspace →
2. 2

   Install your agent

   Pick the coding agent you want to use — Claude Code, Claude Desktop, Cursor, Codex, or Amp. If you already have one installed, skip ahead to step 3. Each agent has a dedicated setup guide under AI & Agents if you need the full install instructions.

3. 3

   Connect your agent to Planwright MCP

   Run the command or add the config for your agent below. OAuth handles authentication — your browser opens for a one-time GitHub login, then the Planwright tools appear in your agent automatically.

   Claude CodeClaude DesktopCursorCodexAmp

   # OAuth — recommended (zero config, browser login)
   claude mcp add planwright --transport sse https://mcp.planwright.tools/sse

   OAuth opens a browser for GitHub login on first use — no token to manage.

4. 4

   Open your agent in any folder

   Create an empty folder and open your agent there. No repository clone required — the demo project has no real GitHub repo. The agent will create source files locally as it implements the objective, guided entirely by the architecture.md context file in Planwright.

   mkdir my-planwright-run && cd my-planwright-run

5. 5

   Give the agent this prompt

   Paste the prompt below into your agent. It connects to the demo project, picks up a scheduled objective, reads the codebase context, implements the work, and requests acceptance — the full Planwright workflow in one session.

   Copy

   Connect to the Planwright demo project and claim a scheduled objective.
   
   Use the MCP tool planwright_set_repo with:
     workspaceSlug: "planwright-demo"
     projectName: "Explore this project"
   
   Then call planwright_list_objectives to see what's available in the Scheduled lane.
   Claim one objective, read the architecture.md context file to understand the codebase,
   implement the objective in this folder, then call planwright_request_acceptance when done.

6. 6

   Watch the board update

   As the agent works, the objective moves through the lanes in real time — Scheduled → In Progress as it claims the work, then In Progress → Acceptance when it calls planwright_request_acceptance. Open the demo board in a browser tab and refresh to see the progress.

7. 7

   Review and accept

   Once the objective reaches the Acceptance lane, click it to open the detail view. You'll see the agent's decomposition plan, a diff summary, and the acceptance criteria. Go through the checklist and click Accept — or Reject with a note if something needs to change. Either way, your decision is written to the audit chain and signed with your identity.

8. 8

   Explore the rest of the app

   You've seen the core loop. Here's what else is in the sidebar.

   Project

   • Board — The kanban view where objectives flow through Backlog → Scheduled → In Progress → Acceptance → Done. Every card is an agent work item — not a task or a ticket. Core Concepts →
   • Initiatives — Group objectives into strategic themes with target dates. Gives the board a planning dimension — you see not just what's being built, but why and by when. Organize with Initiatives →
   • Signals — Per-project metrics: cycle time, acceptance rate, agent throughput, and review velocity. Know whether your directing is working before the quarter ends. Signals & Insights →
   • Bugs — A separate bug lifecycle that never mixes with objectives. Bugs have their own triage, assignment, and resolution flow — they don't pollute the objective board. Core Concepts →
   • Context Files — Markdown files agents read before starting work. Describe your architecture, conventions, or anything an agent needs to orient itself — no README hunting required. Context Files →

   Workspace

   • Release Plan — A visual roadmap of your initiatives with target dates and release status. See what's committed, what's slipping, and what shipped. Release Notes →
   • Changelog — A living record of every shipped objective, organized by initiative and date. Ready to paste into a release email or publish as a public changelog. Release Notes →
   • Insights — Workspace-wide analytics across all projects: strategic focus distribution, team throughput, and agent utilization. The view your CTO wants in the Monday standup. Signals & Insights →
   • Audit Log — The immutable hash-chained record of every action in the workspace — every lane change, agent run, and human sign-off, signed with Ed25519. Required for SOC 2 evidence packages. Core Concepts →
   • Settings — Connect the GitHub App, generate MCP tokens for agents, manage workspace members, and configure billing — all in one place. GitHub Integration →