Shipping the MCP was step one. Now ship the door.
An MCP endpoint nobody can find is an integration, not a channel. The teams winning the early AI-distribution game treat their MCP like a product launch: a page, a directory listing, one-click installs, and an announcement loop.
The distribution playbook
- Your own "/use-from-ai" page. One URL, the three client paths (Claude connector, Cursor one-click, ChatGPT connector), and the hosted connect link. This page also ranks for "<your product> MCP" — a query class growing every month.
- Directory listings. Public MCP directories are becoming the app stores of the agent era. List with your branding and a working connect link; skip anything sensitive.
- One-click installs in announcements. A changelog post that says "now usable from Claude" converts radically better with an install link than with a config snippet.
- Generative-engine surface. Assistants answer "can I use X from here?" from indexed pages. Structured connect pages, FAQ schema, and directory entries are how your product becomes the answer.
The public-surface rules
A listing is marketing surface — treat it with production discipline: expose only name, logo, description and the connect URL; keep descriptions link-free (spam vector); and make sure a paused or deleted MCP disappears from the listing automatically.
Built-in directory + install kit
FAQ
- How do customers discover that my product has an MCP server?
- Four channels that compound: a /mcp or 'Use us from Claude' page on your own site with the connect link; a listing in MCP directories; one-click install links (Cursor deep links, Claude connector URL) everywhere you announce it; and in-product prompts to power users who already use AI tools.
- What should an MCP directory listing contain?
- Name, logo, one-sentence description of what an assistant can do with it, and a connect link — and nothing sensitive: no tokens, no owner contact, no internal URLs. Treat the listing as public marketing surface, because it is.
- Does an MCP listing help SEO?
- Yes, twice: classic search (people query '<product> MCP' and '<product> Claude integration' in growing numbers) and generative engines — when a user asks an assistant 'can I use <product> from here?', indexed, structured listings and connect pages are what the answer is built from.