Launch your SaaS on directories with Codex

What Codex needs

The repository, cloned locally

After purchase you accept a GitHub invitation and clone the private repository. Start Codex in that folder so AGENTS.md, RUNBOOK.md and the playbooks are in its working tree.

Python 3.10+ on macOS or Linux

Workspace creation, record.py and the local dashboard are Python standard library only. Nothing else is installed; Windows is not supported.

An MCP browser server

Codex reaches Chrome through an MCP server such as Chrome DevTools MCP, registered in Codex’s MCP configuration. BROWSER-SETUP.md describes the two connection modes; follow Codex’s own setup UI for where the entry goes.

Permissions recorded once per product

authorizations.json holds what Codex may do — free registrations, final submissions, verification emails, badges. Granted during onboarding, respected on every resume, no spending.

Connecting Codex to your Chrome

BROWSER-SETUP.md starts from one rule: use the browser you can see and supervise, with one long-lived connection per batch. The tracker does not control Chrome; the agent’s MCP integration does.

Preferred route: in the Chrome you already use (144 or newer), enable remote debugging on the inspect page, then register Chrome DevTools MCP with auto-connect in Codex’s MCP configuration. Trigger a harmless browser action from Codex and approve Chrome’s connection prompt in the intended window; confirm the visible tab moves.

Fallback: a dedicated Chrome with its own user-data directory listening on a local debugging port. Sign in to the needed directory, Google and mail accounts in that window and point the MCP server at that endpoint instead of auto-connect. Never combine the two modes and never expose the port.

Smoke check before any submission: blank tab, public directory homepage, read the heading, confirm the tab, then sign in to a target directory and inspect its account menu. Save only pass/fail notes — never transcripts, cookies or credentials.

The start prompt

The public start prompt from START.md, unchanged. Paste it into Codex inside the cloned folder: it creates your private workspace next to the checkout, starts the local dashboard and replies with a 127.0.0.1 link.

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.

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.

The dashboard then leads onboarding — product, identity, permissions, smoke check — and holds the prompts you hand back to Codex for a batch, a resume or a review. Everything is written to files Codex reads on the next turn.

What a run looks like

The illustration below is the home page’s: a small batch, each platform ending in the state the tracker observed. Your real statuses live in submissions.csv; this is not a measurement of Codex.

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

Codex stops, records prepared_needs_human and the dashboard lists the platform under “needs you”. You solve the challenge in the same window; no evasion.

2FA and verification codes

Codes arrive in the mailbox signed in inside the connected browser. Forwarding a code by hand is a supported fallback and is recorded as a human handoff.

Paid upsells

A route that costs money becomes deferred_paid with a note. Codex never buys, never starts a trial, never accepts an unrelated offer.

Confirming “live”

Marking a platform live needs a public URL and your confirmation. A saved form, a receipt or a thank-you page are recorded as what they are.

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

Questions