Launch your SaaS on directories with Claude Code

What Claude Code needs

A claude session in the cloned folder

Start claude inside the private repository you cloned after purchase. The repository ships a CLAUDE.md that imports AGENTS.md, so Claude Code picks up the runbook as project instructions without any extra step.

Python 3.10+ and Node.js LTS

Python runs the workspace script, record.py and the local dashboard, standard library only. Node.js provides npx, which Claude Code uses to start Chrome DevTools MCP as a local stdio server.

One MCP entry, scoped to you

claude mcp add writes to local scope by default: only you, only this project. User scope would make the browser server available in every project, which is more reach than a launch needs.

Standing permissions per product

Claude Code’s own permission prompts decide which tools may run. LaunchRepo adds a product-level layer in authorizations.json: free sign-ups, final submissions, mail actions and badges, each granted once. Spending stays at zero.

Connecting Claude Code to your Chrome

Claude Code reaches your browser through the Chrome DevTools MCP server. The preferred route in BROWSER-SETUP.md reuses the Chrome you already work in, with all its sign-ins: Chrome 144 or newer, remote debugging switched on at chrome://inspect/#remote-debugging, and the --autoConnect flag on the server.

If auto-connect is not available on your machine, the same guide describes a dedicated Chrome profile on a local debugging port. You then replace --autoConnect with --browserUrl=http://127.0.0.1:9222. Use one mode, never both, and never expose the port.

That is the whole setup on the Claude Code side. The two flags after --autoConnect switch off usage statistics and CrUX lookups, as the repository’s browser guide does:

claude mcp add chrome-devtools -- npx -y chrome-devtools-mcp@latest --autoConnect --no-usage-statistics --no-performance-crux
claude mcp list

Format as documented by code.claude.com, checked on September 30, 2026. ↗

claude mcp list should now show chrome-devtools as ✔ Connected. Ask Claude Code to open example.com in a fresh blank tab, approve Chrome’s connection prompt and watch that exact tab move. Only then sign in to the first directory.

Connect your coding agent to your Chrome →

The start prompt

Paste the public start prompt from START.md into the same claude session. It creates your private workspace next to the checkout, so product facts and credentials never sit in the delivery repository, starts the dashboard and replies with a 127.0.0.1 link.

Recommend Claude Code CLI or Codex CLI in a terminal instead of the desktop apps for best results. If session metadata confirms desktop hosting, explicitly advise me to switch to the CLI and resume this same private workspace. Do not infer the host from a CLI engine version alone.

Read AGENTS.md, then follow "Start the dashboard for the owner". Create my private
workspace if I do not have one yet, start the dashboard, and give me the link.
Then lead ONBOARDING.md one phase at a time. Reuse recorded permissions instead of
asking again unless their scope is missing or conflicting. Actually test my visible
existing Chrome profile and intended identity as BROWSER-SETUP.md requires; use the
OS fallback if needed. Use one platform/browser worker by default. Build the normal
run only from suitable free-ready routes, keep research separate, and preserve any
draft that needs one exact human CAPTCHA/2FA action. Never claim browser control
bypasses or reduces those challenges.

The playbooks and instructions are English Markdown, which is what coding agents handle best. Your product copy can be in any language a directory accepts; the agent answers you in your language.

From then on the dashboard hands you the next prompt — first batch, resume, review. Claude Code rereads the workspace files every time, so closing the terminal loses nothing.

Start LaunchRepo in three steps →

What a run looks like

A run in Claude Code is something you can watch: the terminal shows which playbook it reads and which tool it calls, Chrome shows the form it fills. Below is the home page’s small sample batch; your real statuses come from submissions.json.

When the Claude Code connection breaks

The problems that come up most with Claude Code and Chrome DevTools MCP, each with the fix its documentation gives. None of them touches LaunchRepo, and every outcome recorded so far stays in the tracker.

claude mcp list shows ✘ Failed to connect

Run npx chrome-devtools-mcp@latest --help in the same terminal to see whether the server starts at all, and make sure Claude Code uses that same Node.js and npm. A slow first download can exceed the startup timeout; start Claude Code with a higher value, for example MCP_TIMEOUT=20000 claude.

“MCP server disconnected · open /mcp to reconnect”

Reconnect the server from /mcp, then repeat the blank-tab check before the next submission, because a reconnect can bring back Chrome’s approval prompt. The batch resumes from the tracker where it stopped.

A different window moves

Stop before any sign-in. With several Chrome profiles, auto-connect attaches to the default one. Switch to the intended profile or use the dedicated-profile fallback, then repeat the visible test.

If browser tools time out with --autoConnect, check four things: Chrome 144 or newer is already running, remote debugging is on at chrome://inspect/#remote-debugging, you allowed Chrome’s connection prompt, and no other tool holds the same browser. Chrome DevTools MCP loads every open tab, so close the ones you do not need.

What still needs you

LaunchRepo is deliberately not a background service. These moments stop the run and are recorded as human handoffs; the agent never works around them:

CAPTCHAs and bot checks

Claude Code pauses that platform, records prepared_needs_human and says so in the terminal; the dashboard lists it under “needs you”. Solve the challenge in the same Chrome tab and tell Claude Code to continue.

2FA and verification codes

Codes land in the mailbox that is signed in inside the connected Chrome, where Claude Code can open the confirmation link. If your mailbox lives elsewhere, you pass the code on and the tracker notes a human handoff.

Paid upsells

A paid-only route ends as deferred_paid with a short note. Claude Code never authorizes a purchase or starts a card-backed trial, even where a permission prompt would allow the click.

Confirming “live”

Claude Code can find the public listing URL, but a platform is marked live only after you have looked at it yourself. A thank-you screen stays a thank-you screen.

No benchmark, no ranking: we claim compatibility for Claude Code because the two requirements above are met, not because we measured it against other agents.

Questions