Launch your SaaS on directories with Cursor

What Cursor needs

The cloned repository as your workspace

Open the private repository in Cursor after purchase. The agent reads AGENTS.md and the playbooks from the workspace, and you keep product.json and submissions.json open in the editor while it works.

Python 3.10+ and Node.js LTS

Python runs the workspace script, record.py and the dashboard in Cursor’s integrated terminal. Node.js provides npx for the browser server.

A global ~/.cursor/mcp.json

Cursor reads MCP servers from ~/.cursor/mcp.json for every project or from .cursor/mcp.json for one. The global file keeps the delivery repository exactly as it arrived.

Tool approval left on

Cursor asks before each MCP tool call by default. Keep it that way for form submits; authorizations.json adds LaunchRepo’s product-level rules on top.

Connecting Cursor to your Chrome

Cursor’s agent has no way into your signed-in Chrome until you add one. The route LaunchRepo supports is Chrome DevTools MCP over remote debugging: Chrome 144 or newer, remote debugging switched on at chrome://inspect/#remote-debugging, and --autoConnect on the server.

You can add the server under Cursor Settings → MCP → New MCP Server, or paste the JSON below into the file. Either way it appears in the MCP list with its tools, and you can switch it off without deleting it.

The same definition as in the repository’s browser guide, saved as ~/.cursor/mcp.json:

{
  "mcpServers": {
    "chrome-devtools": {
      "command": "npx",
      "args": [
        "-y",
        "chrome-devtools-mcp@latest",
        "--autoConnect",
        "--no-usage-statistics",
        "--no-performance-crux"
      ]
    }
  }
}

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

Ask the agent to open example.com in a fresh blank tab. Approve the tool call in Cursor, allow Chrome’s prompt and check that the visible tab moved. 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 Cursor’s agent with the repository open. It creates the private workspace next to the checkout, starts the dashboard and returns 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.

Onboarding continues in the dashboard. Because the workspace files are plain JSON and Markdown, you can also correct a fact straight in the editor; the agent rereads the files before the next platform.

Start LaunchRepo in three steps →

What a run looks like

In Cursor you follow a run in two places: the agent panel lists every tool call with its arguments, and Chrome shows the page. The panel below is the home page’s sample batch; real statuses come from submissions.json.

When the browser server fails in Cursor

Cursor isolates a failing MCP server, so the rest of the editor keeps working. These three checks usually get the browser back:

The agent does not see the tools

Open the Output panel (Cmd/Ctrl+Shift+U) and choose “MCP Logs”. It shows whether the server started, failed or crashed. A server switched off in Customize does not load at all.

An old server version keeps running

For npm-based servers Cursor’s docs suggest removing the server, running npm cache clean --force and adding it again. Restart Cursor afterwards.

A tool call failed mid-form

Cursor marks the call as failed. Look at the tab before retrying: if the form was not sent, retry; if it was, let the agent record what the page shows instead of submitting twice.

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

The agent stops, records prepared_needs_human and the dashboard lists the platform under “needs you”. Solve it in the Chrome tab, then tell the agent in the panel to carry on.

2FA and verification codes

Confirmation emails reach the mailbox signed in inside the connected Chrome. If it is not there, you pass the code on and it is logged as a human handoff.

Paid upsells

Paid-only routes are recorded as deferred_paid. Even with a tool call approved, the agent does not buy, start a card-backed trial or accept an unrelated offer.

Confirming “live”

You mark a platform live once you have opened its public URL. Until then drafts, receipts and thank-you pages keep their own states.

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

Questions