Boss Agent CLI is a command-line client for BOSS Zhipin, the Chinese recruitment platform, built for both people and AI agents. It covers the job-seeker workflow (job search, benefit filtering, a local shortlist, resume work with AI, applications, and chat) and the recruiter workflow (candidate search, recommendations, greetings, resume requests and downloads, chat, and job management) in one tool.
People run an interactive terminal wizard that walks through role, platform, and goal. Agents call the same workflows through four routes: an MCP server exposing 77 tools for hosts such as Claude Desktop and Cursor, subprocess calls that parse JSON from stdout, a typed Python API, and an OpenCode project example. Every command writes a single JSON envelope with ok, data, pagination, error, and hints fields, and a schema command describes all capabilities and can export them as OpenAI or Anthropic tool definitions.
Benefit filtering is the distinguishing search feature: it pages through results, fetches job details, and matches requested benefits such as weekends off or social insurance with AND logic, then sorts by a local match score. AI features (job-description analysis, resume polishing and targeting, shortlist matching, mock interviews, cover letters, and chat coaching) call any OpenAI-compatible or Anthropic-compatible API, including local models through Ollama or vLLM.
State stays on the local machine: encrypted login credentials, search cache, shortlist, resumes, and AI settings. A resumable crawl mode runs under fixed budgets for requests, details, wall-clock time, and retries, saves SQLite checkpoints, and stops when the platform returns a risk signal. Errors carry a code, a recoverable flag, and a recovery action so agents can respond programmatically. A platform abstraction allows further recruitment sites, though BOSS Zhipin is the only one registered.
Features
- Job search: keyword search with eight filter dimensions and cached results you can revisit by number
- Benefit filtering: auto-paginates and fetches details to match requested benefits with AND logic, sortable by a local match score
- Terminal wizard: an interactive flow for people that agents can also drive and resume by run ID or MCP
- Local shortlist and stats: save, tag, annotate, and compare candidate jobs offline, with funnel statistics
- AI job-search tools: JD analysis, resume polishing and optimization, mock interviews, cover letters, and chat coaching
- Recruiter workflow: candidate search, recommendations, greetings, resume requests and downloads, chat replies, and job management
- JSON envelope: every command returns the same ok, data, pagination, error, and hints shape on stdout
- Agent access: an MCP server with 77 tools, subprocess calls, and a typed Python API
- Tool-definition export: the schema command exports capabilities as OpenAI or Anthropic tool definitions
- Resumable crawls: budgeted batch collection with SQLite checkpoints and JSON, CSV, or XLSX output
- Structured recovery: errors include a code, a recoverable flag, and a recovery action

