Where to Launch an AI Agent: GitHub, Workflow Libraries and Registries
An AI agent earns trust by showing what it did, so launch it where people can read the code or watch a run: a GitHub repository with topics, a workflow library such as n8n templates, the official MCP Registry if it exposes tools, and a Show HN post with a recorded run. Add a free board listing for a public page. Product Hunt suits agents with a clear, visual outcome.
What does an agent launch have to prove that a chatbot launch does not?
A chatbot answers. An agent acts: it sends the email, edits the file, books the slot, spends the credits. People therefore ask a different set of questions. What can it touch? Can I see what it did? What happens when it is wrong at step six of ten? A launch place is good for an agent when it lets you answer those with evidence instead of adjectives.
That points toward places where code, logs and permissions are visible, and away from places that reward a polished tagline. A short list of what to have ready before any listing: the exact permissions it asks for, a recorded run of one real task from start to finish, and one run where it failed and how it recovered.
Where do people who adopt agents actually look?
Platform facts below were read on each platform’s own pages on 13 November 2026, apart from the ones whose sites refused an automated check, which we mark.
01 GitHub
The place most agents are first read, because the repository is the product page.
- Mechanism
- Repository topics, a README and releases
- Found through
- Topics, search and links from other pages
- Reader
- A developer deciding whether to run your code on their machine
GitHub’s docs describe topics as labels that classify a repository so others can find it. Add the ones a reader would search, naming the framework you use and the job the agent does. Then put a recorded run, an architecture sketch and a permissions table at the top of the README, above the install steps.
If the agent runs only as a hosted service, still publish a thin public repository with examples and an issue tracker. A visible place to report failures does more for trust than a claim of reliability.
02 n8n workflow templates
A library of shareable automation workflows, including agent workflows, with a creators directory.
- Route
- Submission through the Creator Dashboard, per its docs
- Quality bar
- Its guide covers requirements, testing and documentation
- Extra
- A verified creator programme with a listing in a creators directory
If your agent can be expressed as a workflow that others import and run on their own n8n instance, this is a distribution channel where users arrive with the tooling already installed. Package one narrow job, such as triaging a support inbox, and document the credentials it needs.
It does not fit an agent that is a standalone application. We read the template guide’s outline but not its review timings, so ask how long approval takes before you plan a launch day around it.
03 Hugging Face Spaces
Demo hosting whose docs also cover exposing a Space as an agent tool or an API endpoint.
- Docs section
- Pages on Spaces as agent tools and Spaces as MCP servers
- Fit
- An agent with a single, safe, sandboxed task
- Avoid
- Agents that need a user’s real accounts
A Space is the right shelf for a read-only agent that researches, summarises or classifies, because a visitor can try it without handing it any permissions. The public page doubles as a demo you can link from your repository.
An agent that writes to someone’s calendar or inbox does not belong in an open demo. Build a sandbox with fake data for it, label it as a sandbox, and link the real install instructions beside it.
04 Official MCP Registry
The official metadata registry for publicly reachable MCP servers, currently in preview.
- Status
- Its docs say it is in preview and that breaking changes or data resets may occur
- Ownership check
- You verify your namespace through GitHub, DNS or HTTP
- Fit
- An agent that offers its abilities to other agents as tools
List in the registry only if other agents are meant to call yours. Then the entry is how clients discover it. For an agent that is used by people through its own interface, the registry is the wrong audience.
The full steps, including the publishing tool and the metadata file, are on the MCP servers page. Here the point is the decision: tool for agents, or product for people?
Visit Official MCP Registry (registry.modelcontextprotocol.io) →
05 Hacker News (Show HN)
A technical audience that will ask what the agent is allowed to do, and then try to break it.
- Cost
- Free to post; account needed
- Guideline
- Something others can try, ideally without sign-up barriers
- Off topic
- Landing pages, newsletters and sign-up pages
The guidelines call off-topic anything that cannot be tried, so a waitlist page for an agent does not qualify. Offer a sandbox or an open-source install, link the recorded run in the post and put the permission list in the first comment.
Expect the first replies to be about safety. Answer them with specifics, such as which actions need a human click, and not reassurance. A prepared list of the guard rails you built will carry the thread.
06 Product Hunt
A general launch site with a daily ranking, suited to agents whose outcome can be shown in a short clip.
- Cost
- Free to submit; sign-in required
- Ranking
- A daily ranking that resets at midnight Pacific
- Access
- Its pages refused our automated check, so only basics are stated
An agent that visibly finishes a chore, like filing receipts into a spreadsheet, makes a good thirty-second clip. One that works invisibly in the background is hard to show and does better on GitHub.
See Product Hunt alternatives for AI agents if you prefer a smaller audience with fewer simultaneous launches.
07 Post your build ideas (our launch board)
A free board we run, with no queue and no account, where an agent gets a public page and a comment thread.
- Cost
- Free; one emailed link publishes it
- Ranking
- Verified first, then newest, with Today, Week and All-time feeds
- Limit
- A small audience; no promise of reach
Use the launch form when you want a stable public page to link from your repository and your posts. Browse the AI agent ideas first to see how other builders phrase what their agents do, and borrow the clarity, not the wording.
A published launch is final, so tune the description before you press Publish, while the questions agent users ask are still guesses; a later correction is an email to support@postyourbuildideas.com. We do not promise visitors; this is a page and a thread, not a traffic source.
Which place matches which kind of agent?
| Agent type | Start with | Then add | Skip for now |
|---|---|---|---|
| Coding or repo agent | GitHub | Show HN | Product Hunt |
| Workflow or ops agent | n8n templates | GitHub | Spaces |
| Research or summary agent | Hugging Face Spaces | Show HN | Registry |
| Tool offered to other agents | MCP Registry | GitHub | Product Hunt |
| Consumer agent with a visible result | Product Hunt | Our board | Registry |
Then read the playbook for launching an AI agent for the dated plan, and the guide to building an app with an AI coding agent if you are still shaping it.
Frequently asked questions
Where should an open-source AI agent launch first?
On GitHub, with topics, a recorded run and a clear README, because that repository is what every other place will link to. Then post to Show HN with the install steps ready. An open-source agent also fits the MCP Registry if it exposes tools, and workflow libraries if it can be exported as one.
Do I need an agent directory to launch an AI agent?
No. General directories rarely separate agents from other AI tools, so a listing there reaches the same visitors as any other. Spend the time on a repository or demo that shows a real run. If a directory has an agents category, read its submit page and confirm whether listing is free.
How do I show that my AI agent is safe to try?
Publish the permissions it requests, run it in a sandbox with fake data for the public demo, and show one failure with the recovery. Say which actions require a human confirmation. Safety claims without these details read as marketing, and technical audiences will ask for them in the first reply.
Post your build ideas publishes this page and runs a free launch board that competes with some of the sites it names. Facts about other sites came from their own pages on the date shown above; terms, queues and prices change, so confirm on their sites before you decide. About the publisher.