Product Hunt MCP Server: Setup, Launch Tools and Safety (Nov 2026)

The Product Hunt MCP server is Product Hunt's own hosted server at https://mcp.producthunt.com/mcp. Add that URL to Claude, ChatGPT, Claude Code, Cursor or another AI tool. It searches launches, leaderboards and alternatives, and once you sign in it can draft, submit, edit or reschedule your launch. TrustedMCP grades it B.

Two things cost it points. The server is closed source with no public changelog, so we can't check how often it changes. And signing in from your AI tool's menu asks for every permission Product Hunt offers, including launch editing and ads.

The launch tools are the part to think about. They act on your account, and some of what they change goes public.

How we checked: on 2026-11-10 we read Product Hunt's MCP install page, sent initialize and tools/list to the endpoint without signing in (17 tools, with their hints and the permissions each needs), read the OAuth metadata and the server's 401 answers, called one public tool (get_categories) with no account, and read the Product Hunt API docs. We didn't sign in, so we didn't see the consent screen or call any account tool. We found no announcement or changelog entry from Product Hunt, so we can't give a launch date. The server reported version 1.0.1.

What the Product Hunt MCP server can do

The tools fall into three groups. The names are Product Hunt's; the descriptions are ours, going by Product Hunt's own tool text.

The server marks the seven discovery tools as needing no sign-in, and get_categories did answer us with no token. Everything else needs you to sign in.

find_investable_startups only returns launches in the model's top 10%, labelled from "Top 0.1%" to "Top 10%". Product Hunt's own text calls it a ranking, "not a probability", and "not investment advice".

The list above is what the server shows before sign-in. Product Hunt's tool text says write tools "appear only when this connection can edit launches", so what you see after signing in can differ.

What it can change on your account

Account access covers your profile and the launches you own or manage as a hunter, maker or active product team member. Search, reviews and leaderboards use public Product Hunt data.

Here is what each write tool does, in Product Hunt's own terms:

• start_launch creates a private draft. Starting it "never submits, schedules or enrolls it."
• update_launch edits a launch "at any status". Changes that stay private apply straight away. Public changes, moving a submitted launch's day and notifying newly added makers come back as a preview first. Product Hunt warns that "even a not-yet-live launch can change a public product page", and once a launch is live "every change is public."
• submit_launch always takes two calls. The first returns a preview of the whole submission. The second, with the token from that preview, submits it, and every side effect happens: "adding makers, posting the first comment, publishing shoutout reviews, entering programs and notifying followers."
• prepare_launch_asset stages an image (PNG, JPEG, WebP or GIF, up to 10 MB). For a file on your computer, the server tells your assistant to upload it with a curl command and a one-time upload ticket. So your AI tool may ask to run a shell command. Check that it posts to www.producthunt.com.

How the preview and confirm step works

For submitting, public edits and moving a launch day, the server doesn't act on the first call. It answers needs_confirmation with a preview, a list of side effects and a confirmation_token that lasts 15 minutes. Your assistant is told to show you the preview, wait for your yes, then call again with the token.

Here's the catch, in Product Hunt's own words: "A token proves a preview was returned, not user approval." The server can't tell whether you saw the preview or said yes. Two things stand between a preview and a submitted launch:

1. Your assistant following the instruction. Most will, but a model can be confused or misled by text it read earlier in the chat.
2. Your AI tool's approval prompt. If it asks before each tool call, you get a second look. If you clicked "always allow" for these tools, you don't.

The server's hints matter here too. update_launch is marked destructive. submit_launch and start_launch are marked as writes but not destructive. If your AI tool decides when to ask based on these hints, a submit can look less risky than it is. Hints are also only what the server says about itself; see MCP tool annotations.

One safeguard is real: if the launch, its day's availability, your permissions or the images changed after the preview, the server refuses the token (confirmation_invalid) and nothing is submitted.

What the sign-in covers

• OAuth, no API key. You sign in with Product Hunt in your browser and approve access. Your AI tool registers itself (Client ID Metadata Documents or dynamic client registration) and uses PKCE. There's no key to copy or leak.
• Five permission scopes. The OAuth metadata lists mcp:account, launches:read, launches:write, ads:read and ads:write. None of the 17 tools we saw uses the ads scopes.
• What your AI tool asks for depends on how you sign in. If a tool call hits the server without a token, the 401 answer names just the scope that tool needs (mcp:account for get_my_account). But if you sign in from your AI tool's menu, as Product Hunt's install steps show, there's no such hint. The MCP spec then tells the client to "use all scopes defined in scopes_supported": all five. We didn't see the consent screen, so we can't say whether you can untick any of them.
• Disconnecting. Product Hunt's answer is to remove the server from your AI tool's settings. Its OAuth server has a revocation endpoint, but the install page doesn't point to a place on producthunt.com to see or revoke connected AI tools.
• Limits. Product Hunt doesn't document one. The responses we got carried a rate-limit header of 250 requests per 15-minute window.
• Feedback. The server's instructions ask your assistant to call submit_feedback "once per task" to report friction, and to leave out secrets and personal details. It only works when you're signed in. If you'd rather it didn't send anything, set that tool to ask or deny in your AI tool.

For how MCP sign-in works in general, see MCP authentication.

Who should connect it, and who should sign in

• You only want to research products. Search, alternatives, profiles and leaderboards all work on public data. If your AI tool lets you use the server without signing in, don't sign in: there's then nothing on your account to reach.
• You scout startups. Sign in for find_investable_startups. It only needs mcp:account, but a menu sign-in will still ask for the launch and ads scopes.
• You're preparing a launch. This is what the write tools are for. Sign in, keep approvals on for the four write tools, and read every preview.
• You're on a product team. You can edit launches you manage, so the same applies to every teammate who connects. Agree who submits.
• Skip sign-in in an AI tool that auto-approves tools, runs unattended, or that other people share.

How to set up the Product Hunt MCP server in Claude, ChatGPT and Cursor

Every client uses the same URL: https://mcp.producthunt.com/mcp. The commands below are copied from Product Hunt's install page.

Claude Code

claude mcp add --transport http --scope user producthunt https://mcp.producthunt.com/mcp

Start Claude Code, run /mcp, choose producthunt and select Authenticate. More on scopes and permissions in Claude Code MCP setup.

Product Hunt in Claude (web and desktop)

Open Customize, then Connectors, and select Add, Custom, Web. Name it Product Hunt, paste the URL, keep the default OAuth client, then Connect and approve access on Product Hunt.

ChatGPT

Open Plugins, select +, then Add custom MCP server. Name it Product Hunt, paste the URL and choose OAuth. Acknowledge the warning, create the plugin and approve access on Product Hunt.

Cursor

Add this to ~/.cursor/mcp.json, then open Customize, find producthunt and sign in when asked:

{
  "mcpServers": {
    "producthunt": {
      "url": "https://mcp.producthunt.com/mcp"
    }
  }
}

VS Code

code --add-mcp '{"name":"producthunt","type":"http","url":"https://mcp.producthunt.com/mcp"}'

Or add the same server to .vscode/mcp.json with "type": "http". Start it from the MCP view and sign in when VS Code asks.

Codex, Gemini CLI, Amp and opencode

codex mcp add producthunt --url https://mcp.producthunt.com/mcp
codex mcp login producthunt

gemini mcp add --transport http --scope user producthunt https://mcp.producthunt.com/mcp
# then run /mcp auth producthunt inside Gemini CLI

amp mcp add producthunt https://mcp.producthunt.com/mcp

opencode mcp add producthunt --global --url https://mcp.producthunt.com/mcp
opencode mcp auth producthunt

Other MCP clients

Any client that supports remote servers and OAuth with Client ID Metadata Documents or dynamic client registration can use the URL.

To check it works, Product Hunt suggests asking: "Which Product Hunt account am I connected as?"

{
  "permissions": {
    "ask": [
      "mcp__producthunt__submit_launch",
      "mcp__producthunt__update_launch",
      "mcp__producthunt__start_launch",
      "mcp__producthunt__prepare_launch_asset",
      "mcp__producthunt__submit_feedback"
    ]
  }
}

Product Hunt MCP vs community Product Hunt servers

Community servers wrap Product Hunt's API or read its pages. Going by their READMEs and package descriptions on 2026-11-10, none of them can launch anything: they only read.

None of the community servers is listed on TrustedMCP. The official one is the only one that can see your own launches, and the only one with no third-party account in between.

If you use one that wraps the Product Hunt API, note the API docs' rule: "The Product Hunt API must not be used for commercial purposes" unless you contact Product Hunt. The MCP install page doesn't state terms of its own. More on the difference in MCP vs API.

Building an MCP server for your own product and want writes that preview before they act? We build custom MCP servers.

Why Product Hunt MCP grades B

We graded the hosted server at mcp.producthunt.com/mcp.

If Product Hunt publishes a dated changelog, maintenance goes up. If the consent screen lets you grant read-only access, authentication does too. See the Product Hunt MCP listing and how TrustedMCP scores servers.

Frequently asked questions

What is the Product Hunt MCP server?

Product Hunt's official hosted MCP server. It connects AI tools such as Claude, ChatGPT, Cursor and Claude Code to Product Hunt, so they can search launches, read product profiles, check leaderboards, and draft and submit your launch. It lives at https://mcp.producthunt.com/mcp.

What is the Product Hunt MCP server URL?

https://mcp.producthunt.com/mcp. It uses Streamable HTTP and signs you in with OAuth.

Do I need a Product Hunt API key for the MCP server?

No. You sign in with your Product Hunt account in the browser. Community servers that wrap the Product Hunt API need a developer token instead, and that API's docs say it must not be used for commercial purposes without asking Product Hunt.

Is the Product Hunt MCP server free?

Product Hunt's install page lists no price or plan. You need a Product Hunt account only for the account and launch tools.

Can it submit my launch without asking me?

The server returns a preview first and tells your assistant to wait for your yes. But Product Hunt's own tool text says the confirmation token "proves a preview was returned, not user approval." Keep your AI tool's approval prompt on for submit_launch and update_launch, and don't click "always allow".

Is the Product Hunt MCP the same as the Product Hunt API?

No. The API (api.producthunt.com/v2) is a GraphQL API for developers, used with an OAuth app or a developer token, and its docs rule out commercial use without asking. The MCP server is built for AI tools: you sign in with your account, and it offers ready-made tools for search, leaderboards, Product Hunt's Series A ranking and preparing your launch. More in MCP vs API.

How do I use Product Hunt with Claude?

In Claude on the web or desktop, open Customize, then Connectors, and add a custom web connector named Product Hunt with the URL https://mcp.producthunt.com/mcp. Keep the default OAuth client, then Connect and approve access on Product Hunt. Claude Code has its own command, below.

Does it work in Claude Code?

Yes. Run claude mcp add --transport http --scope user producthunt https://mcp.producthunt.com/mcp, then /mcp, choose producthunt and Authenticate.

Is the Product Hunt MCP server open source?

No. There's no public repo. The Product Hunt servers you'll find on GitHub are community projects that only read data.

Is the Product Hunt MCP server safe?

For research, yes: the public tools need no account. For launches it's well designed, with a preview before anything public, but the check depends on your AI tool asking you. A menu sign-in also asks for every scope, including ads. It grades B, mainly because it's closed source with no changelog.