MCP server

Q: How is this different from scraping the site?

Don't scrape. Use https://www.scrolllaunch.com/api/ai for a bounded cached snapshot, https://www.scrolllaunch.com/api/mcp for tool calls, or any page's .md twin / Accept: text/markdown. Those surfaces are rate-limited and designed for agents.

Q: Where should I start as an agent?

Read https://www.scrolllaunch.com/llms.txt first, then hit /api/ai for this week's leaderboard, then use MCP tools for search and product detail.

Public, read-only tools for AI agents - search live products, weekly + directories DR leaderboards, product detail by slug. No API key.

Open endpoint

Connect agents to the live catalog

Point any MCP client at our HTTP transport. The same four tools are registered in-browser via WebMCP when the site loads.

POST /api/mcp Read llms.txt AI snapshot

www.scrolllaunch.com/api/mcp

Public tools

API key

None

Access

Read-only

Transport

HTTP

Connect

Endpoints

Point your MCP client at the HTTP transport. The server card is the machine-readable discovery document.

MCP transport

POST JSON-RPC 2.0 · GET returns a short discovery stub

/api/mcp

MCP discovery

Streamable HTTP endpoint + SEP-1960 manifest

/.well-known/mcp.json

Server card

SEP-1649 card - Streamable HTTP, tools, capabilities

/.well-known/mcp/server-card.json

AI discovery snapshot

Bounded JSON: this week's board + recently live

/api/ai

Tools

Four public tools

Same catalog as the server card and the in-browser WebMCP registration.

search_products

Keyword search across the live product directory.

Args: query (required), limit 1–25 (default 10)

get_weekly_leaderboard

Ranked launches for an ISO week (defaults to the current week).

Args: week? (YYYY-Www), limit 1–50 (default 20)

get_product

Full public metadata for one product by slug.

Args: slug (required)

get_directories_dr_leaderboard

Highest-DR indexable startup directories (Ahrefs Domain Rating).

Args: limit 1–100 (default 25)

Examples

JSON-RPC over HTTP

Stateless - no sessions, no SSE. Send one JSON-RPC object per POST.

initialize

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "initialize",
  "params": {
    "protocolVersion": "2024-11-05",
    "capabilities": {},
    "clientInfo": { "name": "example-agent", "version": "1.0.0" }
  }
}

tools/call - search_products

{
  "jsonrpc": "2.0",
  "id": 2,
  "method": "tools/call",
  "params": {
    "name": "search_products",
    "arguments": { "query": "ai saas", "limit": 5 }
  }
}

Also supported: ping, tools/list. Rate-limited; public catalog data only.

FAQ

Common questions

Do I need an API key?

No. The MCP endpoint and /api/ai snapshot are public and read-only. Mutations (submit a product, upvote) still require a signed-in user - see auth.md.

How is this different from scraping the site?

Don't scrape. Use /api/ai for a bounded cached snapshot, /api/mcp for tool calls, or any page's .md twin / Accept: text/markdown. Those surfaces are rate-limited and designed for agents.

Where should I start as an agent?

Read llms.txt first, then hit /api/ai for this week's leaderboard, then use MCP tools for search and product detail.

Start with the lightest surface

Agents: read llms.txt, grab /api/ai for a bounded snapshot, then call MCP tools for search and detail. Prefer HTTP for server-side agents; WebMCP mirrors the same tools in-browser.

Read llms.txt Open MCP endpoint