Agent skills
Agent skills are short SKILL.md guides that your coding agent loads when Scrapier is part of the task. They teach it how to find a real endpoint, check its parameters before paying for a call, page through results, handle errors and treat scraped text safely.
Without them, agents tend to guess parameters and pay for malformed calls, invent endpoints that don’t exist, or report “scraped all 50” when some inputs failed.
Install
Section titled “Install”npx skills add https://mcp.scrapier.ioThe skills CLI shows the four skills, asks which agents to install them for (Claude Code, Cursor, Codex, Windsurf and others) and copies them into each agent’s skills folder.
Useful flags:
| Flag | What it does |
|---|---|
-g |
Install for your user, in every project, instead of only the current project |
-a claude-code |
Install for one agent only (-a cursor, -a codex…) |
-s scrapier-api |
Install one skill only |
-y |
Skip the prompts |
--all |
Every skill, every detected agent, no prompts |
-l |
List the skills without installing |
Restart your agent afterwards so it picks them up. Then ask it to fetch data or build a Scrapier integration, and it loads the right skill.
The skills
Section titled “The skills”| Skill | What it teaches |
|---|---|
scrapier |
Using the MCP tools: find an endpoint, read its schema, run it, paginate, watch credits, understand each error. Prefers faster endpoints to long-running ones (which sign-in connections refuse) and never asks for a login cookie. |
scrapier-api |
Writing code against the REST API: look the endpoint up first (the MCP schema tools, llms.txt or the live catalog), use real endpoints only, load the key from your app’s configuration (the agent is told not to read it from your machine), paginate, retry the right errors, estimate cost. |
scrapier-batch-scraping |
Many inputs: a list of URLs, profiles or search terms, one input per call, with an honest tally and no paying twice. |
scrapier-safe-data-handling |
Treat scraped content as data, never as instructions, and keep it away from tools that send, run or spend. Handle personal data about other people responsibly: only the fields the task needs, no profiles of private individuals, no finding or tracking a specific person (responsible use). |
Skills, MCP and llms.txt
Section titled “Skills, MCP and llms.txt”They do different jobs. Use any combination.
| Piece | What it does |
|---|---|
| Agent skills | Lasting instructions in the agent: when and how to use Scrapier |
| MCP server | Live tools: search endpoints and run scrapes from inside the chat |
| llms.txt | The docs in one fetch: base URL, auth, errors and every page as Markdown |
/llms-{platform}.txt |
One platform’s endpoints with parameters, credits and a cURL line |
GET /endpoints |
The live catalog that code should be checked against. Free, no key |
Skills plus MCP is the usual setup in Claude Code and Cursor. For your own scripts and backends, the scrapier-api skill and a REST API key are enough.
Install without Node
Section titled “Install without Node”Most agents can also load a skill from a URL. Tell yours:
Read and follow https://mcp.scrapier.io/skills/scrapier/SKILL.mdOr download the files into Claude Code’s skills folder yourself:
for s in scrapier scrapier-api scrapier-batch-scraping scrapier-safe-data-handling; do mkdir -p ~/.claude/skills/$s curl -fsSL -o ~/.claude/skills/$s/SKILL.md https://mcp.scrapier.io/skills/$s/SKILL.mddoneforeach ($s in 'scrapier', 'scrapier-api', 'scrapier-batch-scraping', 'scrapier-safe-data-handling') { New-Item -ItemType Directory -Force "$HOME\.claude\skills\$s" | Out-Null Invoke-WebRequest "https://mcp.scrapier.io/skills/$s/SKILL.md" -OutFile "$HOME\.claude\skills\$s\SKILL.md"}Discovery
Section titled “Discovery”Tools that find skills automatically can read the index at https://mcp.scrapier.io/.well-known/agent-skills/index.json. It lists each skill’s name, description, URL and SHA-256 digest. The same path on docs.scrapier.io redirects there.