Liepin Recommend MCP

Liepin Recommend MCP is a Node.js MCP/CLI helper for researching and operating Liepin recommend and chat workflows through a Chrome DevTools debugging port.

Install

npm install -g @reconcrap/liepin-mcp
liepin-mcp install --agent trae-cn

install will:

• bootstrap runtime folders and screening-config.json template,
• merge MCP server config into detected external agent config files (Cursor/Trae/Trae-CN/Claude/OpenClaw),
• mirror bundled skills into detected external skills directories.

Usage

liepin-mcp doctor --fix --target-page recommend
liepin-mcp install
liepin-mcp install --agent openclaw
liepin-mcp self-heal --target-page recommend --provider-check
liepin-mcp skill export --format markdown
liepin-mcp external-agent config
liepin-mcp research discover --debug-port 9222
liepin-mcp recommend start --candidate-limit 5 --scan-limit 10 --recommend-criteria "推荐筛选条件" --chat-criteria "聊天筛选条件"
liepin-mcp chat start --candidate-limit all --job "全部职位" --unread-only false --criteria "聊天筛选条件"
liepin-mcp search start --profile "测试" --job "招聘实习生" --hide-read true --candidate-limit 5 --criteria "搜索筛选条件"
liepin-mcp recommend-chat start --candidate-limit 5 --scan-limit 10 --recommend-criteria "推荐筛选条件" --chat-criteria "聊天筛选条件"
liepin-mcp runs progress --kind chat --include-completed false

Doctor/start preflight will automatically handle fixable environment issues: install missing npm dependencies, open Chrome with the configured remote debugging port, and navigate to the target Liepin page (recommend, search, or chat). The operator is only needed for issues that cannot be solved safely by automation, such as Liepin login, captcha/risk pages, or filling real LLM config values.

recommend/chat/recommend-chat start now default to production behavior:

• real recommend chat clicks are enabled by default,
• real request-resume clicks are enabled by default.

For chat start, --candidate-limit accepts a positive integer or all-candidates expressions such as all, 全部, 所有, and 扫到底; all-candidates mode scans until the chat list bottom/platform limit.

Use --execute-request-resume false or a dry-run workflow when you need no-side-effect validation.

Use the MCP tool liepin_run_progress to query progress across recommend, search, chat, and recommend-chat runs. Pass run_id for one run, or omit it to list recent runs with optional kind, limit, and include_completed filters.

Long-run robustness

Long-run recovery is enabled by default in the official release:

liepin-mcp recommend-chat start --candidate-limit 5 --scan-limit 10
liepin-mcp search start --profile "测试" --job "招聘实习生" --hide-read true --candidate-limit 5 --criteria "搜索筛选条件"
liepin-mcp chat start --candidate-limit all --job "全部职位" --unread-only false --criteria "聊天筛选条件"

Modes:

• recover: default long-run behavior with heartbeat, timing, additive checkpoint metadata, and bounded safe-boundary recovery.
• observe: records heartbeat, candidate timing, phase timing, and additive checkpoint metadata.
• off: rollback-safe legacy behavior.

Rollback is immediate: start the next run with --robustness-mode off or pin the package back to @reconcrap/[email protected].

Safety Notes

• The automation is designed to preserve list state and avoid refreshes or hard navigation when returning from details.
• Browser automation is background-safe and does not call Page.bringToFront.
• Real outreach or resume-request actions should only be run with explicit operator approval.

Development

npm install
npm test