goofish-cli
MCP server for automating Xianyu (Goofish) marketplace operations: publish, manage items, handle IM messages, and more via CLI or AI agents.
README
<div align="center">
goofish-cli
闲鱼 CLI · 原生支持 MCP · 为 AI Agent 而生
Goofish (Xianyu) automation CLI · MCP-ready · Built for AI Agents
</div>
goofish-cli 把闲鱼(Xianyu/Goofish)的核心运营能力抽成一套结构化命令,
同一份定义同时输出给三种消费者:
- 👨💻 人类:
goofish item get 12345 --format table - 🤖 AI Agent(Claude Code / Cursor / Codex):
uvx goofish-cli→ 自动注册成 MCP tool - 🧩 Claude Skills(v0.3):5 个内置 skill,
goofish skills install一行装到~/.claude/skills/
架构思想来自 opencli 的 single-registry 设计。
✨ 核心特性
- 🔐 16 个命令覆盖核心链路:发布、下架、查询、图片上传、AI 类目识别、默认地址、IM 收发 + 会话列表、skills 安装
- 📡 真·实时 IM:WebSocket 长连 + 自动重连 + 三类事件分类输出
event=message(收到消息)·event=read(已读回执)·event=new_msg(轻量通知)
- 🛡 内置风控护栏:令牌桶限流(1 写/分钟)+ RGV587 自动熔断
- 🧠 AI-first I/O:
--format json/yaml/table/md/csv,给 LLM 喂 JSON、给人看表格 - ⚡ 一次定义,三种入口:CLI / MCP / Skill 共享同一 registry
- ✅ 真实端到端验证:每个命令都跑过真实账号
🚀 60 秒上手
# 1. 安装
pip install goofish-cli # 或 uv pip install goofish-cli
# 2. 导入 cookie(从浏览器 DevTools → Application → Cookies 导出)
goofish auth login ~/Downloads/goofish-cookies.json
# 3. 验证登录态
goofish auth status
# → {"unb":"2214350705775","tracknick":"xy575986224572","nick":"...","valid":true}
# 4. 干活
goofish item get 1045171414271
goofish message watch # 实时接收消息
goofish message send <cid> <toid> --text "在的" # 发消息
🧩 Claude Skills(v0.3 新增)
v0.3 起内置 5 个 Claude Skill,装完之后 Claude Code / Cursor 里的 Agent 在识别到 闲鱼任务就会自动加载对应 skill 的知识库,不再靠 zero-shot 试错。
# 装到 ~/.claude/skills/(默认)
goofish skills install
# 或者只看有哪些 skill,不拷贝
goofish skills install --list
# 或者装到自定义目录 + 覆盖已有的
goofish skills install --dest ./skills --force
5 个 skill 分工
| Skill | 什么时候激活 | 核心能力 |
|---|---|---|
goofish-overview |
用户首次提闲鱼 / 问这工具能干啥 | 总入口,dispatch 到其他 4 个 skill |
goofish-publish-item |
发商品 / 上架 / 挂闲置 | 类目识别 → 标题 5 段式 → 风控扫描 → 图片检查 → 确认发布 |
goofish-reply-buyer |
回消息 / 看未读 / 议价 | 拉未读 → 意图 5 分类 → 议价三档(小刀/大刀/屠龙刀)→ 用户确认发送 |
goofish-risk-guard |
发布前 / 发送前 / 被限流了 | 违禁词表、外联词正则、发布红线、x5sec 恢复指引(被其他 skill 频繁引用) |
goofish-shop-diagnosis |
店铺没流量 / 曝光掉了 | 买家视角搜索 + 历史对比 → 归因清单 + 修复建议(纯读不写) |
skill 的源文件在仓库的 skills/ 目录下(每个 skill 一个子目录,含 SKILL.md
references/*.md知识库)。也可以走 Claude Code Plugin Marketplace 安装:
claude /plugin marketplace add fancyboi999/goofish-cli
📟 命令详略与真实输出
<details open> <summary><b><code>goofish list-commands</code></b> — 注册表全景</summary>
$ goofish list-commands --format table
| 命令 | 说明 | 写操作 |
|---|---|---|
auth login |
从 JSON 文件或 cookie 字符串导入登录态 | ❌ |
auth status |
检查登录态是否有效 | ❌ |
auth reset-guard |
手动解除风控熔断 | ❌ |
item get |
查询闲鱼商品详情 | ❌ |
item publish |
发布商品(自动识别类目 + 默认地址) | ✅ |
item delete |
下架/删除商品 | ✅ |
media upload |
上传图片到闲鱼 CDN | ✅ |
category recommend |
AI 识别商品类目 | ❌ |
location default |
获取默认发布地址 | ❌ |
message list-chats |
拉取会话列表(左栏;--watch-secs N 叠加 WS 历史推送补漏) |
❌ |
search items |
搜索闲鱼商品(浏览器路径 Playwright + 系统 Chrome) | ❌ |
item view |
浏览器视角看商品详情(字段完整,抗风控;item get 的姊妹版) |
❌ |
message history |
拉取会话历史消息 | ❌ |
message send |
发送文本/图片 | ✅ |
message watch |
常驻 IM 长连(JSONL 输出) | ❌ |
</details>
<details> <summary><b><code>goofish auth status</code></b> — 登录态健康检查</summary>
{
"unb": "2214350705775",
"tracknick": "xy575986224572",
"nick": "闲鱼用户昵称",
"valid": true,
"h5_token_exp": "2026-04-21T20:30:00+08:00"
}
</details>
<details> <summary><b><code>goofish message watch</code></b> — 三类事件 JSONL 流</summary>
$ goofish message watch
实时输出(小号给主号发 3 条 + 主号读了所有消息):
{"event":"message","cid":"60585751957","send_user_id":"2215266653893","send_user_name":"小号昵称","send_message":"测试消息1"}
{"event":"message","cid":"60585751957","send_user_id":"2215266653893","send_user_name":"小号昵称","send_message":"测试消息2"}
{"event":"message","cid":"60585751957","send_user_id":"2215266653893","send_user_name":"小号昵称","send_message":"测试消息3"}
{"event":"read","cid":"60585751957","msg_ids":["4077151826249.PNM","4066820235744.PNM","4066826134477.PNM"],"status":1,"ts":"1776770953455"}
| 事件 | 字段 |
|---|---|
message |
cid · send_user_id · send_user_name · send_message · content_type |
read |
cid · msg_ids[] · status · ts |
new_msg |
cid · msg_id · ts(服务端只推指针,需 message history 拉正文) |
自动跳过噪音:/s/para(对方正在输入)、contentType=8(会话激活心跳)。
</details>
<details> <summary><b><code>goofish message send</code></b> — 主动发消息</summary>
$ goofish message send 60585751957 2215266653893 \
--text "在的 claude 测试成功 ✅" --item-id 1045171414271
{"ok": true, "mid": "1061776769407570", "cid": "60585751957"}
</details>
<details> <summary><b><code>goofish item publish</code></b> — 发布商品(含风控护栏)</summary>
$ goofish item publish \
--title "男士毛呢大衣 驼色长款" \
--desc "全新未拆封 原价 2999 现 999" \
--images ./a.png,./b.png \
--price 999
流程:
media upload每张图 → CDN URL + 尺寸category recommend拿 AI 识别的 catIdlocation default拿默认地址mtop.idle.pc.idleitem.publish落库
返回:
{"ok": true, "itemId": "1046118265141", "status": "published"}
触发令牌桶限流(1 写/分钟)。高频调用会被本地拒绝,避免被闲鱼风控。 </details>
🔌 接入 Claude Code(MCP)
在 ~/.config/claude-code/config.json:
{
"mcpServers": {
"goofish": {
"command": "uvx",
"args": ["goofish-cli"]
}
}
}
Claude 会自动把全部命令看成 tool:goofish_item_get / goofish_item_publish / goofish_message_watch... 你在对话里直接说"帮我看下 itemId=xxx 的详情",Claude 就会调用。
🎯 项目亮点
| 能力 | 说明 |
|---|---|
| 11 个核心 mtop 接口 | 发布/下架/查询/图片/类目/地址/IM 全覆盖 |
CLI + --format 多格式输出 |
json / yaml / table / md / csv,人机两用 |
| MCP Server | uvx goofish-cli 一行接入 Claude Code / Cursor |
| WebSocket 批量 push 全量解码 | 一帧多条消息全部还原,不丢单 |
| WebSocket 自动重连 | 断线自退避重连,长跑无感知 |
| 已读回执 / typing / 新消息通知分类 | /s/sync 元事件结构化为三类 JSONL |
| 全局限流 + 风控熔断 | 令牌桶 1 写/分钟 + RGV587 自动熔断 |
| 单元测试 | 33 个,ruff 零告警 |
| 包分发 | pip install goofish-cli / uvx goofish-cli |
🗺 Roadmap
- [x] v0.1:12 个命令 + MCP + IM 三类事件
- [x] v0.2:
goofish message list-chats(会话列表 + sessionType 分类:1 真人 / 3 系统 / 6 互动 / 23 通知;--watch-secs支持合并 WSackDiff(pts=0)历史推送补齐 h5 接口漏掉的会话) - [x] v0.2:浏览器自动化链路(吸纳 OpenCLI 精华)—— Playwright + 系统 Chrome 驱动
goofish search items/goofish item view,抗风控 & 完整字段 - [x] v0.2.3 / v0.2.4:session 自动续命(passport 快速进入)+
auth login --qr扫码兜底 - [x] v0.3:Claude Skills 包装(5 个 skill:overview / risk-guard / publish-item / reply-buyer / shop-diagnosis)+
goofish skills install - [ ] v0.4:
goofish message create-chat(主动与陌生用户建会话) - [ ] v0.4:
goofish order(订单状态查询 / 发货) - [ ] v0.4:历史数据落盘(SQLite / JSONL),给 shop-diagnosis 做时序归因
- [ ] v0.5:支持发视频消息
🛠 开发
git clone https://github.com/fancyboi999/goofish-cli
cd goofish-cli
uv venv --python 3.11
uv pip install -e ".[dev]"
uv run pytest # 33 测全绿
uv run ruff check src tests # 零告警
详细请看 CONTRIBUTING.md 和 docs/architecture.md。
⚠️ 合规声明
本工具仅用于用户自有账号的自动化运营。严禁:
- 欺诈 / 刷单 / 虚假交易
- 针对闲鱼平台的 SaaS 化转售
- 违反闲鱼、淘宝、阿里巴巴用户协议的行为
工具不提供:绕过滑块验证、批量设备 ID 伪造、自动化规避封号。遇到风控请人工处理(见 docs/compliance.md)。
📜 License
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.