Agent skills for web scraping,
and the scraper stays yours

An agent skill is a folder with a SKILL.md inside, and the agent reads that file when a prompt matches what it says it covers. This one carries the HasData endpoints, their parameters and the shape of what comes back, so the agent writes a working scraper instead of guessing. Install it once with npx skills add hasdata/agent-skills, then ask for the data you want in plain language.

MCP gives the agent tools to call during a conversation, and every page it fetches lands in the context window. A skill gives it knowledge instead, and what you keep is code in your repository that runs afterwards without an agent. On a run of a few hundred pages that difference is the whole cost, because the scraper writes rows to disk while the agent stays out of the loop. Same endpoints, same balance, different thing to own.

Yes. They are markdown in a public repository, so you can install them, read them and fork them. Credits only move when the code your agent writes calls the API, and you start with 1,000 free, no card.

One line covers most of them. npx skills add hasdata/agent-skills detects what you have installed and writes into each agent's own skills directory, and that registry already knows Claude Code, Cursor, VS Code through Copilot, Codex, Gemini CLI, Devin, Zed and Warp among others. Anything else that reads a SKILL.md can point at the two files directly.

Only for the hasdata-cli skill, which drives terminal work. The hasdata skill needs nothing but your key.

One on the free tier, so a loop over pages runs sequentially. Paid plans start at 15 concurrent requests. Past your limit the API answers 429, and nothing is charged for the calls that failed.

No. It activates when a prompt looks like a HasData job, whether that is scraping, SERP, lead enrichment or price tracking, and the CLI skill even carries a section named Non-obvious triggers for the cases where nobody says "scrape". /hasdata and /hasdata-cli call it explicitly.

Whichever your project already uses. The skill carries endpoints, parameters and response shapes rather than a client library, so the agent writes the request in your stack.

You can, it's markdown. What this one saves is the part that goes stale. Which endpoint answers a given task, what its parameters are called, what comes back. Re-running the add pulls whatever the repo has, so the maintenance is one line instead of a rewrite.