Launch your AI agent product on 346 directories with your own agent

The specific pain: everyone lists agents, nobody agrees on the fields

Agent products land somewhere between “AI tool”, “developer tool” and “SaaS”, and the directories reflect that confusion:

Which category is an agent?

AI tool directories, agent-specific lists and general launch platforms all take agent products, each with different category trees and different questions about models and autonomy.

Autonomy claims get you rejected

Directories ask what the agent can do. Overstating it is the fastest route to a removed listing; understating it hides the product.

Your own agent needs guardrails, not just access

Pointing an agent at fifty forms without stop conditions produces duplicates, accepted upsells and “done” that means nothing.

How the workflow fits an agent builder

1. 01

   A runbook with explicit stop conditions

   AGENTS.md and RUNBOOK.md tell the agent when to search for duplicates, when to confirm the signed-in identity, when to decline an upsell and when to stop and ask. Your product gets the guardrails you would want for your own agents.

2. 02

   Facts, not autonomy marketing

   product.json separates shipped features from planned ones. The agent copies from the facts and never upgrades “planned” to “available” to fit a field.

3. 03

   Any agent that reads files and drives Chrome

   Claude Code, Codex, Cursor or a browser agent; the same setup guide covers Chrome DevTools MCP and a dedicated CDP profile. You can even use the agent you are launching, if it meets those two requirements.

4. 04

   Outcomes with evidence

   record.py accepts eleven states, each tied to a minimum observation. “Live” needs a public URL and your confirmation; a completed tool call is never proof that a page saved.

Which directory categories matter

The catalog tags every one of the 346 platforms with one category. The counts below are computed from the catalog at build time; how many platforms actually fit your product is something your agent checks per directory before it submits anything.

• AI tool directories

  100 playbooks

  The AI tool directories, including the agent-specific lists in the catalog. Where an agent product is most clearly at home.

• Product launch platforms

  104 playbooks

  General launch platforms for the launch day and the first wave beyond the AI niche.

• Communities

  7 playbooks

  Builder communities where agent products get discussed rather than listed. The agent prepares; you post.

Example platforms from the catalog

Example platforms from the catalog that accept agent products. Two have a public field guide on this site; the others are described in the repository playbooks.

• Stork.AIAn AI tool directory that analyses your URL first; the free route needs its badge in your homepage HTML.
• ToolPilotAn AI tool directory with an embedded form, tight word limits and a free-with-badge route.
• Product HuntThe launch day agent products still plan around; the playbook covers scheduling and the maker checklist.Field guide →
• PeerlistA builder community whose Launchpad takes a project from your personal profile into a weekly slot.
• UneedA daily launch platform; the free plan allows one pending launch at a time, so a waiting product is not a failed one.Field guide →

What LaunchRepo does not do

• It is not an agent and ships no model. It is the input your agent needs: playbooks, product context, runbook and tracker.
• It does not guarantee listings, backlinks or traffic. Directories moderate as they like, and some only publish paid spots.
• It does not bypass CAPTCHAs, bot checks or 2FA. Those stop the run and are handed to you; no stealth, no evasion.
• It does not authorise spending. Paid routes are recorded as deferred_paid; your agent never buys anything.

Licence, in one sentence: pay once, then use it for an unlimited number of software products that you or your company own and operate, with as many agents as you like. Launching products for clients needs a separate licence.

Questions