Submit your app from Claude Code: the MCP way to launch

You built the app in a coding agent. The listing should not send you back to a browser tab and twelve form fields. A handful of app directories now run MCP servers, and against one of those your agent can search the catalogue, list the places worth submitting to, and open a draft listing from your URL, all from the prompt you are already typing in.

This guide is the practical version: what MCP is in one paragraph, how to connect a directory in one command, the prompts that work, and the line an agent should never cross.

MCP in one paragraph

The Model Context Protocol is a small standard for giving a model tools. A server publishes a list of things it can do, each with a name and a typed input. A client, which is your coding agent, reads that list and calls a tool when the conversation needs it. Claude Code, Claude Desktop and Cursor all speak it. A directory that runs an MCP server is offering its search, its listings and its submission form as tools rather than as pages.

For a founder that means one thing: the directory can be worked from the terminal. No copying a tagline into a field, no hunting for the submit button, no wondering whether the site accepted the URL.

Connect a directory in one command

Every MCP server is one address. Ours is at spacerrapps.com/api/mcp, and connecting it from Claude Code is a single line:

claude mcp add --transport http spacerrapps https://spacerrapps.com/api/mcp --header "Authorization: Bearer <your token>"

Claude Desktop and Cursor take the same address in their MCP settings as JSON:

{
  "mcpServers": {
    "spacerrapps": {
      "url": "https://spacerrapps.com/api/mcp",
      "headers": { "Authorization": "Bearer <your token>" }
    }
  }
}

The token comes from your account page and is shown once. Reading tools need no token at all, so an agent can search the directory and list the pool before you have signed up for anything. The token is only for the one tool that writes on your behalf.

The prompts that work

Once connected, the tools show up in the agent's list and you talk to it normally. These five get the most done.

Find where else to submit. "List the directories on SpacerrApps with a domain rating above 60 that do not charge, and tell me which three I should do first for a developer tool." The agent calls the directories tool, reads the numbers, and reasons over them.

Check if you are already listed. "Search SpacerrApps for Kettle and tell me if a listing exists." A duplicate submission is the most common reason a directory refuses one.

Open the draft. "Submit https://kettle.app to SpacerrApps and give me the link to finish it." The agent reads your landing page, takes the name, tagline and description it already states, saves a draft on your account and returns the link.

Override what the page says. "Submit https://kettle.app with the tagline 'Focus timing that stays out of your way'." Anything you give the agent wins over what it read.

Write the copy first. "Read https://kettle.app and draft three taglines under sixty characters and a 300 character description for a directory listing." Do this before the submit, paste the one you like into the override, and the draft arrives finished.

Where the agent stops, and why that is the feature

An agent that can open a draft could, in principle, publish it, pay for a placement and pick a launch week. It should not, and a directory that lets it should worry you.

The reason is simple. A listing is a claim about your product, in your name, on a page that will outrank your own site for some queries. The name, the price, the platforms and the category are facts, and a model that fills a field it was not given will fill it with something plausible. Plausible is the failure mode. So the good design is the one where the agent does the typing and you do the deciding: it opens the draft, you check every line in the browser, and a person at the directory reads it before it goes live.

That is how our server works. The one writing tool opens a draft on your account and hands back the address where you finish it. It cannot publish, cannot pay, cannot touch a listing that is already live, and a revoked token stops it at once.

What we learned trying to automate the rest

We spent two days on the other approach: a browser agent that would take your answers and file them on a hundred directories unattended. It listed a few. It also met CAPTCHAs, sign-up forms that only accept a company-domain email, magic links that open an onboarding rather than confirm an account, and fields no founder would want a machine to guess, like a school name on a hackathon site. Every one of those is a place where an unattended bot either stops or invents.

The honest conclusion was that the browser is the wrong tool for the writing half and the right tool for the reviewing half. A draft you finish in thirty seconds beats a bot that submits something you did not read. If a service promises to file your app on a hundred sites while you sleep, ask what it types when a form asks a question you never answered.

A checklist for an agent-native launch

• Put an llms.txt at the root of your site, so an agent reading your page finds a clean summary first. There is a guide to that here.
• Make sure your landing page states the name, a one-line tagline and a real description in plain text. The draft is only as good as what the page says.
• Connect the directory once, from the command above, with a token you can revoke.
• Ask for the draft, open the link, and read every field as if a stranger wrote it.
• Add the logo and screenshots yourself. No agent should pick the picture of your product.
• Keep the review link. It is the same draft the wizard saves, so you can come back to it.

Which directories speak MCP today

Few, and the list is growing. SpacerrApps runs the server described above. Nick Launches publishes one with search, a readiness audit and a draft tool. Most of the larger directories do not yet, and their forms remain forms. When one you care about adds a server, the shape will be the same: open reads, one guarded write, and a link back to a page where you take over.

If you want to try it now, the agents page on this site has the commands ready to copy, and a token takes one click on your account page.