Your score went down. Here's how to move it up.

The single biggest score lever. An MCP server makes your tool callable by Claude Desktop, Cursor, Windsurf, Zed, and every IDE that speaks the protocol. If your product is API-backed, the wrapper takes a weekend. Stork will list it.

Anthropic MCP registry. Cursor's tools list. Claude Desktop config repos. ChatGPT GPT store if relevant. The score rewards presence, not just intent. Stork itself counts — being listed here is one of the surfaces we check.

Per-seat pricing dies when agents replace seats. If your only tier is $X per user per month, agents either can't buy or won't. Add a metered tier — per call, per token, per request. Even $0.01 minimums signal you're built for the new shape.

Self-serve API keys, a sandbox tier that doesn't require sales contact, and rate limits that don't immediately throttle agentic workloads. If a developer has to email "hello@" to get an API key, you fail this check.

Drop an openapi.json at a standard path. Even a rough one is better than none. Agents lean on it to call you safely; LLMs use it to generate correct request shapes without hallucinating.

Active changelog with something shipped in the last 90 days. Silence reads as abandonment — to humans, to agents, and to us. A monthly "what we shipped" post is enough.

This used to be worth five points. As of methodology v1.1 it's worth zero. It takes ten minutes, which makes it trivially gameable, and the independent measurement is unflattering — a 137,000-domain study found 97% of published llms.txt files received zero requests in a month, and no major crawler has confirmed it reads one. We still probe for it, so if that ever changes we'll have the history. Ship it because it's cheap, not because it will move your score or get you cited. The six moves above are the ones that change outcomes.