Valendata

Valendata, from Valendata Inc., saves website tasks as reusable skills that return clean, structured rows and repair themselves when the site changes. Users describe what they need or do it once in their browser, then call the result on demand, on a schedule, from code, or from an AI assistant. The company also publishes an MIT-licensed agent kit on GitHub for Claude Code, Codex and Cursor; the platform that runs skills is a hosted service.

What It Is

Valendata is a web data extraction and automation platform built around skills. A skill is a saved website task with typed inputs and outputs, exposed as its own API endpoint with a cURL call and an MCP address. Skills can be built from a plain-language description or by recording a task once, and a public marketplace lets users clone existing skills into their own editable copy.

How skills stay working

Each run is compared with earlier ones. When data stops coming back, the skill identifies which step broke, re-reads the live page and repairs itself, and a repair goes live only once it brings the data back. Every version is saved so it can be rolled back. Each skill keeps a run log with rows, time and cost, and shows whether it is healthy. Where a site blocks automated browsers entirely, Valendata states that rather than returning made-up rows.

Using it from assistants, code and workflows

Valendata runs an MCP server that works with Claude, ChatGPT, Cursor, Muse and other MCP clients, exposing skills and workflows as tools. The agent kit adds search, run and build capabilities for Claude Code, Codex and Cursor, plus a CLI and web tools to scrape pages to Markdown, crawl, map sites and search. The README describes Firecrawl-compatible and Apify-compatible API endpoints. Workflows chain skills with code, AI steps and connected apps such as Gmail, Slack, Google Sheets and Notion, via Composio. Batch runs accept up to 10,000 inputs, and logins can be saved and are stored encrypted.