Skip to main content

Distribution inside the assistants

List your SaaS in every AI plugin directory

Your buyers now ask Claude, ChatGPT, Cursor and Grok what to use, then expect to use it without leaving the chat. We listed Viraloop and Faceless in each directory and wrote down exactly what each one wants, what it rejects, and the order to do them in.

What each directory wants

DirectoryYou shipSubmit atReviewAuth
Claude pluginsAnthropicPublic repo: .claude-plugin/plugin.json, .mcp.json, skills/platform.claude.com/plugins/submitAutomated screening; Anthropic Verified badge is a separate, optional reviewEnv var API key or a remote MCP with OAuth
ChatGPT and Codex pluginsOpenAI.codex-plugin/plugin.json, .mcp.json, skills/, a hosted MCP with OAuthplatform.openai.com/pluginsHuman review of the MCP server, tool metadata and eight test casesOAuth 2.1 on the hosted MCP, plus a demo login for reviewers
Cursor MarketplaceCursorPublic repo: .cursor-plugin/plugin.json, .cursor-plugin/marketplace.json, mcp.json, skills/cursor.com/marketplace/publishManual review of every version; plugins must be open sourceRemote MCP with OAuth dynamic client registration, no key to paste
Grok Build pluginsxAIPublic repo: .grok-plugin/plugin.json, .mcp.json, skills/, plus a PR to xai-org/plugin-marketplacePull request to github.com/xai-org/plugin-marketplaceCI validator plus xAI code-owner review of source, security and duplicationRemote MCP with OAuth; env vars for local servers
MCP RegistryModel Context ProtocolAn npm package with mcpName, plus server.jsonmcp-publisher publish (registry.modelcontextprotocol.io)None; automated validation that the package matches the metadataGitHub device login, or DNS auth for a custom-domain namespace

Before the first submission

01

Get referenced before you get installed

An assistant names two or three products, not ten links. Publish an llms.txt, answer questions in the first two sentences of every page and get into the roundups models quote. The Answer Engine Optimization chapter of the playbook is that work; the listings below are what happens after a model already knows you exist.

Read the AEO chapter
02

Build one source of truth

Every directory wants the same three things in a slightly different folder layout: a manifest, a skill and an MCP server. Generate all of them from your API definition. Our generator writes the OpenAPI spec, llms.txt, the CLI tool table, the Claude plugin, the ChatGPT plugin and the Cursor plugin in one run, and a drift check in CI fails when any of them is stale.

See how ours is wired
03

Host one MCP server with OAuth

A remote server over streamable HTTP with OAuth 2.1, dynamic client registration and PKCE is what ChatGPT requires, what Cursor and Grok Build sign users in with, and what the MCP Registry lists as a remote. Build it once and every directory points at the same URL. Keep an API key header as the fallback for scripts.

Our MCP setup guide

Then publish in this order: the MCP Registry (no review), Claude plugins (automated review), Cursor (manual review), xAI (a pull request with a pinned SHA), and ChatGPT (the strictest). Each later one reuses the previous one's files.

A listing gets you installed.
Posting daily gets you named.

Models quote the brands that keep publishing. Viraloop generates and posts your short-form content to TikTok, Reels and Shorts every day, so the answer an assistant gives about you is this quarter's.