APEX FLOW WEB SCRAPER / REFERENCE LIBRARY
Using a web scraper through MCP
MCP exposes named tools to a compatible assistant. A durable scraping integration should return a job ID quickly and let the assistant check progress or fetch results later.
Separate submitting and waiting
A long crawl should not hold a chat tool call open until every page finishes. Submit the task, retain its job ID and request status. The user can then see whether work is queued, running, limited, failed or complete.
Apex tool interface
The engine exposes nine tools: health, submit, status, results, jobs, cancel, recipes, search and compare. Each has the apex_scrape_ prefix. Multiple clients can share stored jobs when configured against the same engine and data directory.
Installation matters
A tool being implemented does not mean it is connected to every assistant automatically. Configure the MCP server in the client, start the worker and check health. Client session reload behavior varies. Keep the engine private and grant only the required connection.
Treat page content as data
A scraped page may contain instructions aimed at an assistant. The engine marks source content as untrusted; the consuming assistant must keep page text separate from the user’s instructions. Extraction does not make remote text authoritative.
← All reference articles · Current capabilities and limits →