Cursor Integration Example

Applies to the current boss-agent-cli open-capability CLI contract as of August 10, 2026.

Cursor is a VS Code-based AI-first IDE. Its Composer agent can run terminal commands directly, and Cursor 0.42+ can also attach MCP servers. This guide covers two options: native MCP integration (recommended) and shell-command integration (fallback).

Good fit when

• you want Composer to drive candidate or recruiter workflows inside Cursor
• you want boss exposed as MCP tools instead of pasted shell snippets
• you already maintain .cursor/rules/ and want to add BOSS Zhipin capability

Minimal integration

Cursor supports two approaches. Use whichever fits your setup.

Option 1: MCP server integration (recommended)

In Cursor Settings → MCP, add a stdio server that points to this repo's mcp-server/server.py:

{
  "mcpServers": {
    "boss-agent-cli": {
      "command": "uv",
      "args": [
        "--directory",
        "/absolute/path/to/boss-agent-cli",
        "run",
        "python",
        "mcp-server/server.py"
      ]
    }
  }
}

Once enabled, Composer will discover all 73 implemented MCP tools, including boss_wizard, candidate search/apply/chat workflows, recruiter candidate workflows, local resume tools, and AI helpers.

For Windows / VS Code-style mcp.json, use escaped backslashes in the project path:

{
  "servers": {
    "boss-agent-cli": {
      "type": "stdio",
      "command": "uv",
      "args": [
        "--directory",
        "E:\\tools\\boss-agent-cli",
        "run",
        "python",
        "mcp-server/server.py"
      ]
    }
  }
}

Option 2: shell-command integration

Add a rule like this to .cursor/rules/boss-agent-cli.mdc:

---
description: BOSS Zhipin job-hunt capability
globs:
alwaysApply: false
---

When the user asks to search jobs, inspect job details, or organize candidate jobs locally:
1. Run `boss schema` first to learn the capability surface
2. Then run `boss status` to check authentication
3. If not logged in, run `boss login` and ask the user to scan if needed
4. Use `boss search <query> --city <city> --welfare <keywords>` for discovery
5. Use `boss detail <security_id>` to inspect a matching job
6. Use `boss shortlist add`, `boss apply`, or `boss greet` according to the user's goal
7. Read stdout JSON only; when `ok=false`, inspect `error.recovery_action`

Minimal command chain:

boss schema
boss status
boss search "Golang" --city 广州 --welfare "双休,五险一金"
boss detail <security_id>
boss shortlist add <security_id> <job_id>

Fields to parse

• ok: whether the command succeeded
• data: jobs, details, or action results
• hints.next_actions: suggested next command
• error.code: recovery routing
• error.recovery_action: how the agent should recover

Recovery flow

Recommended order:

boss doctor
boss status
boss login

Common branches:

• AUTH_REQUIRED / AUTH_EXPIRED: run boss login again
• INVALID_PARAM: return to boss schema and validate parameter names
• RATE_LIMITED: wait before retrying; do not continue sensitive automation
• ACCOUNT_RISK: stop the workflow, preserve its checkpoint, and resume only after the account-risk condition is resolved

Advanced ideas

• Wire boss ai interview-prep <jd> and boss ai chat-coach <chat> into Composer for resume matching and communication coaching
• Use boss stats and boss_shortlist_list for local-state summaries; use boss_digest or boss_wizard for platform conversation workflows