Gemini CLI marketing usually means drafting copy, research or content in the terminal. Here it means distribution: listing a SaaS on directories and launch platforms one by one, with your own accounts. Each Gemini CLI fact below links to its docs or repository and carries the date we checked it; where a human has to step in, the page says so.
Gemini CLI features that matter for a launch
We went through the docs and the repository on September 29, 2026. New versions ship often, so every fact links to the page it comes from.
- What it is
- An open-source AI agent from Google that runs in your terminal, under the Apache 2.0 license. Built-in tools cover file operations, shell commands, web fetching and Google Search grounding.github.com ↗ · Checked on September 29, 2026
- Release pace
- The latest stable release when we checked was v0.61.0, published on September 23, 2026. Preview and nightly builds come out in between.github.com ↗ · Checked on September 29, 2026
- Free quota
- Signing in with a personal Google account (Gemini Code Assist for individuals) allows up to 1,000 model requests per user per day. Per-minute limits and service availability also apply, and Gemini CLI picks which Gemini model serves each request.geminicli.com ↗ · Checked on September 29, 2026
- Context files
- Gemini CLI loads GEMINI.md files as persistent instructions: a global one in ~/.gemini, then files in the workspace and its parent folders. The context.fileName setting can add other names, and the docs use AGENTS.md as their example.geminicli.com ↗ · Checked on September 29, 2026
- MCP servers
- MCP servers go into the mcpServers object in settings.json, or are added with gemini mcp add. Unless a server is marked trusted, Gemini CLI asks before each of its tools runs.geminicli.com ↗ · Checked on September 29, 2026
- Browser agent
- A built-in browser agent, off by default, navigates pages, clicks and fills forms through the accessibility tree. It uses a bundled Chrome DevTools MCP server and needs Chrome 144 or newer.geminicli.com ↗ · Checked on September 29, 2026
- Attaching to your Chrome
- In the “existing” session mode the browser agent attaches to a Chrome that is already running, after you enable remote debugging at chrome://inspect/#remote-debugging. Form filling always needs your confirmation, whatever the approval mode.geminicli.com ↗ · Checked on September 29, 2026
- Approval modes
- The --approval-mode flag accepts default, auto_edit, yolo and plan. yolo auto-approves all actions; the older --yolo flag is deprecated.geminicli.com ↗ · Checked on September 29, 2026
- Headless mode
- With -p (or --prompt), or in a non-TTY environment, Gemini CLI runs non-interactively and can return plain text, JSON or a stream of JSON events.geminicli.com ↗ · Checked on September 29, 2026
Where Gemini CLI stops and waits for you
A terminal agent cannot click through a login challenge on your behalf, and directories are designed to make sure of it. These are the stops to plan for:
- Form confirmations
- The built-in browser agent asks you to confirm every form fill, whatever the approval mode. For submissions that is a feature: nothing goes out without your click.
- Sign-ins
- Log in to each directory yourself in the Chrome that Gemini CLI attaches to. Google popups and new-device checks are yours, and passwords never go into a prompt.
- CAPTCHA and bot checks
- Cloudflare, reCAPTCHA and hCaptcha appear on many submit pages. You solve them in the same tab; LaunchRepo never tries to evade one.
- Email codes
- Confirmation links go to your inbox. If it is not open in the attached Chrome, you pass the code on yourself.
- Quota and “live”
- Retries on a stubborn form eat into the daily request quota, and a thank-you page is not a listing. Someone has to record the real outcome and the date of the next check.
GEMINI.md, extensions and launch context
Gemini CLI extensions and prompt packs for marketing mostly cover copy, SEO and social posts. Handy for the launch text, silent on which directories still accept free submissions, which fields a form rejects and whether a listing went public.
Whatever you add to settings.json, an MCP server or an extension, runs with your permissions. Leave trust off for servers you did not write, so every tool call still asks first.
LaunchRepo is not an extension either. The repository holds a runbook in AGENTS.md, 299 Markdown playbooks, one product template and the tracker. Add AGENTS.md to context.fileName and Gemini CLI loads it the way it loads GEMINI.md.
Running LaunchRepo with Gemini CLI
Gemini CLI has two browser routes: its built-in browser agent in “existing” session mode, or Chrome DevTools MCP added as an ordinary server. LaunchRepo’s guide is written for the second, which keeps your setup the same as with any other agent. We have not published a test run with Gemini CLI; this is a match against its docs.
- Load AGENTS.md
- Clone the private repository after purchase, start gemini in that folder and add AGENTS.md to context.fileName, or ask it to read the file first.
- Trust the folder
- Gemini CLI starts stdio MCP servers only in trusted folders; elsewhere they show as Disconnected. Run gemini trust in the cloned repository once.
- Add the browser server
- Merge the entry below into ~/.gemini/settings.json, or use gemini mcp add with -s user. Keep trust set to false.
- Allow Python, set the rules
- The shell tool runs the Python 3.10+ scripts after your approval. authorizations.json then records what Gemini CLI may do for each product, with spending fixed at zero.
The mcpServers block for ~/.gemini/settings.json. It is the server definition from LaunchRepo’s browser guide plus Gemini’s trust flag:
{
"mcpServers": {
"chrome-devtools": {
"command": "npx",
"args": [
"-y",
"chrome-devtools-mcp@latest",
"--autoConnect",
"--no-usage-statistics",
"--no-performance-crux"
],
"trust": false
}
}
}Format as documented by geminicli.com, checked on September 30, 2026. ↗
Run gemini mcp list, or /mcp inside a session: chrome-devtools should read CONNECTED. Then ask for example.com in a blank tab, allow Chrome’s prompt and confirm that the visible tab moved.
Connect your coding agent to your Chrome →
Now the public start prompt from START.md, word for word. Do not be put off by its first line recommending Claude Code CLI or Codex CLI; that is our general default, and every later instruction applies to Gemini CLI unchanged.
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. Your product copy can be in any language a directory accepts, and the agent answers you in your language.
Start LaunchRepo in three steps →
LaunchRepo with Claude Code →
When the Gemini CLI connection fails
What the Gemini CLI and Chrome DevTools MCP docs suggest, in the order that solves most cases:
- chrome-devtools shows Disconnected
- gemini mcp list only starts and tests stdio servers in trusted folders. Run gemini trust in the repository and list again.
- The server will not start at all
- Run npx chrome-devtools-mcp@latest --help in the same terminal to check Node.js and npm. ERR_MODULE_NOT_FOUND points to an unsupported Node version or a damaged npx cache; the Chrome DevTools MCP troubleshooting guide explains how to clear it.
- The quota runs out mid-batch
- Stop, let record.py note what each platform currently shows, and continue after the reset. The tracker, not the chat, holds the state of the run.
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.
Splitting a launch between Gemini CLI, LaunchRepo and you
Five launch tasks, three ways of handling them.
| Task | Gemini CLI alone | With LaunchRepo | You |
|---|---|---|---|
| Launch copy | Writes it in the terminal | Pulls each field from product.json | Sign off on the facts |
| Platform list | Google Search grounding, unverified | Chooses from 299 dated playbooks | Decide the batch size |
| Form rules | Learns them on each page | Reads them from the playbook | Fill a gap when asked |
| Challenges | Waits for your input | Logs a human handoff, continues elsewhere | Clear it in the same tab |
| Outcome record | Gone when the session ends | record.py writes submissions.json | Check the public URL |
Common mistakes
- Submitting in yolo or headless mode
- Auto-approval and -p are fine for reading files or checking the tracker. Submissions need you nearby for sign-ins, challenges, form confirmations and Chrome’s prompt.
- Using the browser agent’s default profile
- In “persistent” mode the built-in browser agent opens its own Chrome profile with none of your accounts signed in. Attach to your own Chrome instead.
- Burning the daily quota on retries
- The free tier counts model requests, and a long form with many retries uses them fast. Work in small batches and stop when a platform keeps failing.
- Trusting the browser server
- The --trust flag skips every confirmation prompt for that server. For a tool that can press submit, keep the prompts.
Questions
Can Gemini CLI control a browser?
Yes. Its built-in browser agent uses Chrome DevTools MCP and needs Chrome 144 or newer; you can also add Chrome DevTools MCP yourself as an MCP server. In either case it can attach to a Chrome you already run, once remote debugging is on.
Is Gemini CLI free?
The tool is open source. With a personal Google account the documented free quota is up to 1,000 model requests per day, with per-minute limits on top. Paid Google plans and API keys come with their own limits.
Does Gemini CLI read AGENTS.md?
By default it looks for GEMINI.md. Set context.fileName in settings.json to include AGENTS.md, as the docs show, or ask it to read the file at the start of a session.
Can Gemini CLI do marketing for my SaaS?
Parts of it, yes: drafting copy, researching platforms and filling in forms in an attached browser. For a directory launch it has no per-platform rules and keeps no record of results between sessions, and sign-ins, CAPTCHAs and 2FA stay with you.
Can Gemini CLI solve CAPTCHAs?
Nothing in its docs says so, and LaunchRepo does not try. A challenge becomes a recorded human handoff: you solve it in the same window, then the agent continues.
What does it cost?
Gemini CLI is free to install; the free quota or your Google plan covers model usage. LaunchRepo costs €119.00 once at the current price, for unlimited own products and any agent.
Give your agent the launch playbooks
The current price is €119.00, paid once, for unlimited own products and any agent. The price does not rise.
See pricing