ppxc-leads-mcp

找客户 · Find Customers｜销售线索与评论区获客

找客户 · Find Customers｜销售线索与评论区获客（品牌别名：OPC 评论线索雷达）是一款 Agent Skill / MCP 工具，帮助商家从抖音、小红书、快手公开评论中识别购买意向、销售线索和可跟进客户名单。

找客户 · Find Customers｜销售线索与评论区获客 (brand alias: OPC Comment Lead Radar) is an Agent Skill and MCP tool for lead generation from short-video comments, helping teams discover customer intent and sales leads from Douyin, Xiaohongshu, and Kuaishou.

Find Customers lets an MCP-capable assistant turn public comments on supported social platforms into customer signals, ranked leads, follow-up scripts, and reusable customer-pool records.

Package: https://www.npmjs.com/package/ppxc-leads-mcp
Setup guide: https://opc1.me/download/mcp Comment lead generation use case: https://opc1.me/use-cases/comment-lead-generation Comment lead radar vs. export tools: https://opc1.me/compare/comment-lead-radar-vs-export-tools

Maintainer Notes

本机最新 MCP 源码目录是 /Users/jianshi/ppxc-mcp。ppxc-leads-mcp 是 npm 包名、命令名和远端仓库名，不是当前本机源码目录名。

更新 MCP 工具本体、工具返回、搜索流程、进度通知、Skill 工作流或 npm 包发布前，先在仓库根目录运行：

npm run guard:mcp-target

检查不通过时不要继续构建、发包或验证。/Users/jianshi/ppxc 只用于后端和官网支撑层，不是 MCP 主源码目录。

What It Does

• Finds high-intent sales leads from public comments.
• Detects purchase intent, comparison questions, complaints, and recommendation requests.
• Generates follow-up scripts that a sales or private-traffic team can use.
• Saves results into the user's customer pool after OPC login and save/unlock.
• Generates online HTML reports from OPC backend database records instead of local files or Blob-first storage.
• Helps content teams turn real customer questions into the next content angles.

Best-fit searches: lead generation, sales leads, social media leads, comment analysis, intent detection, customer discovery, CRM, social listening, local business leads, Xiaohongshu leads, Douyin leads, Kuaishou leads.

Install MCP

Supported desktop systems:

• macOS with Node.js 18 or newer.
• Windows 10/11 with Node.js 18 or newer. Use the Windows config below when the host cannot launch npx directly.

Add this server to an MCP-capable host:

{
  "mcpServers": {
    "ppxc-leads": {
      "command": "npx",
      "args": ["-y", "ppxc-leads-mcp"]
    }
  }
}

Recommended Windows config:

{
  "mcpServers": {
    "ppxc-leads": {
      "command": "cmd",
      "args": ["/c", "npx", "-y", "ppxc-leads-mcp"]
    }
  }
}

First launch downloads the local browser runtime. On Windows this happens under the user's profile and can take one or two minutes depending on network access to the Electron mirror. If the host reports that the connector exited immediately, retry with the Windows cmd /c config above before changing anything else.

Windows Support Status

Windows is an intended desktop runtime path for this package. Current coverage:

• Packaged MCP smoke runs on GitHub Actions windows-latest.
• The launcher uses a Windows-specific stdin pipe so Electron does not exit immediately under GUI process semantics.
• Platform login and verification windows include Windows focus recovery.

Windows real-machine light smoke and one low-frequency full E2E partial-pass have been recorded. Remaining requirement before saying "Windows full official support": run a non-empty lead case that saves/claims at least one lead and queries it back from the customer pool. See docs/windows-official-support.md.

Install Skill

The official workflow skill is bundled in the same npm package. After the MCP connector is enabled, tell the assistant:

Check and install the companion Find Customers Skill for this host.

The assistant should call setup_companion_skill with action=status first. If the Skill is missing or outdated, it must show the fixed target directory and ask for your approval. Only after you approve may it call action=install with the current host target and confirm=true. Refresh or restart the host after a successful install.

The installer supports the standard local Skill roots for Codex, Cursor, Claude, Agent Skills, and WorkBuddy. It never runs from npm postinstall, never writes to an arbitrary path, and does not silently install or update anything. Cloud hosts that do not expose a local Skill directory should use the public Skill marketplace or plugin entry instead.

Public companion Skill entries:

• GitHub source and Claude/Gemini plugin: https://github.com/yuanjian068yuan/opc-comment-lead-radar
• SkillHub: https://skillhub.cloud.tencent.com/skills/ppxc-lead-radar-20260620
• ClawHub: https://clawhub.ai/yuanjian068yuan/opc-comment-lead-radar
• skills.sh: https://www.skills.sh/yuanjian068yuan/opc-comment-lead-radar/find-customers

Manual fallback: the bundled workflow is located at:

skills/ppxc-find-customers/SKILL.md

Install the whole skills/ppxc-find-customers/ folder into your assistant's skill directory, or import the folder through the host's skill UI. For Codex Desktop / Codex CLI, the usual destination is:

~/.codex/skills/ppxc-find-customers/

MCP installation exposes the tools. Skill installation teaches the assistant the safe workflow: read the workflow manifest, ask for the product/service and target platform, run a trial analysis first, show the first two customer signals, then ask the user to log in only when they want to save, unlock, or query the full customer pool. OPC login is not required before a trial scan.

Tools

The package exposes seventeen MCP tools:

• setup_companion_skill
• check_status_and_login
• get_workflow_manifest
• list_products
• suggest_search_keywords
• start_search_run
• get_search_run_status
• start_batch_search_run
• get_batch_search_run_status
• cancel_current_search_run
• search_keyword_for_leads
• analyze_video_comments
• query_leads
• mark_lead_feedback
• update_lead_status
• review_followup_queue
• export_diagnostics

Privacy And Safety

• Platform login sessions and OPC login credentials stay on the user's machine.
• Platform tokens and cookies are not uploaded to OPC and are not written to diagnostics.
• Public comment text is sent to OPC over HTTPS for customer-intent analysis, follow-up suggestions, and customer-pool storage.
• Trial scans show the first two complete customer signals and do not save to the customer pool.
• Saved results are stored only in the user's own OPC customer pool.
• Online reports for saved searches are rendered by the OPC backend with owner checks.
• Captchas, QR-code login, and security checks are handled by the user in a local window. This package does not bypass verification.
• The tool does not send private messages or automate outreach.

Troubleshooting

If the assistant says the MCP tools are unavailable, restart or refresh the MCP host after adding the config.

If the MCP tools are available but the assistant does not know the full workflow, ask it to call setup_companion_skill with action=status. Approve installation only after it shows the expected host and Skill directory.

If startup is slow, wait for the browser runtime download to finish. In mainland China networks, the setup guide includes a mirror configuration for Electron downloads.

If a platform asks for QR-code login or verification, complete it manually in the local window, then retry the assistant request.

If something fails, ask the assistant to run export_diagnostics and send the generated diagnostic file to OPC support.

License

This package is proprietary. It may be installed and used as part of the OPC service, but copying, modifying, reverse engineering, republishing, or redistributing it is not allowed except with written permission from OPC.