Skip to main content
April 24, 2026
MCP Server
Weekly update

New features

  • Dual-mode execute_task — The execute_task tool now automatically picks the best execution path for your client. Clients that support MCP Task receive real-time progress updates and streamed results as the run progresses. Other clients get a task ID immediately and can poll for results, so long-running scrapes no longer block the session.
  • API key authentication — You can now connect to the Octoparse MCP Server using an API key in addition to OAuth. API keys make setup easier in headless or automated environments such as Claude Code, VS Code, TRAE, and OpenClaw. Generate a key from the Octoparse account center.

Updates

  • Free trial credits for MCP — Free and Basic accounts now receive 2,000 records per week for MCP extraction, with no upgrade required to get started. See Prerequisites.
  • Gemini support — Gemini is now a fully supported client for the MCP Server. See the setup guide.
April 3, 2026
CLI
Weekly update

New features

  • octoparse detect — New command that opens the Octoparse extension browser, inspects a URL, and generates a local task file. Supports --auto for automatic region selection, --manual for guided browser interaction, and --agent / prepare-preview-apply for AI agent workflows. See Create tasks from a URL.
  • OAuth loginoctoparse auth login --oauth opens a browser-based OAuth flow and stores access and refresh tokens locally. Useful when you prefer not to manage API keys manually.
  • auth info command — New octoparse auth info subcommand shows detailed account and credential information for the active session.
  • OCTO_ENGINE_ACCESS_TOKEN — New environment variable for bearer token authentication. Useful in short-lived CI jobs where storing credentials is not desirable.

Updates

  • JSONL event types expandedrun --jsonl now emits captcha, proxy, download.started, download.succeeded, and download.failed events in addition to the previous set. See Output and exit codes.
  • --lot-id flag for exportdata export and local export now support --lot-id <lotId> to export a specific run lot rather than the latest.
  • --chrome-path for diagnosticsdoctor and browser doctor now accept --chrome-path to test a specific Chrome installation.
  • Linux arm64 clarified — Local extraction (run, detect) is not supported on Linux arm64 because Chrome for Testing has no Linux arm64 package. Cloud extraction works on all platforms.