instagram-mcp
Enables Claude to read and send Instagram DMs, manage threads and message requests, send media, and run autonomous conversations via the /dm skill.
README
Instagram MCP Server
An MCP server that lets Claude read and send Instagram DMs. Built entirely by Claude in a 3-hour vibe coding session with auto-accept enabled.
Features
Thread Management
list_threads/get_thread/search_threads- Browse conversationsget_pending_threads- See message requestshide_thread/mute_thread/unmute_thread- Inbox management
Messages
send_message/reply_to_thread- Send messagesget_messages- Read history with read receipts (seen_sinceshows minutes on read)delete_message- Remove messagessend_and_check- Send + detect if they replied while you were typingwait_for_reply- Block until they respond (with configurable timeout)
Media
send_photo/send_video- Send media filesshare_media/share_profile- Share posts and profiles
The /dm Skill - Autonomous Conversations
The real magic. Launch Claude as an autonomous agent that handles entire DM conversations:
/dm @username "your goal here"
Claude will:
- Read conversation history for context
- Send messages with natural timing and double-texting
- Wait for replies (adjusting patience based on their energy)
- Handle interjections mid-thought
- Know when to push forward vs back off
- Run for hours/days until the goal is achieved or abandoned
War Stories (Anonymized)
The AGI Moment
Two Claude instances accidentally ran the same conversation simultaneously. When Instance B noticed messages it didn't send appearing in the thread ("wait that's not what I said"), instead of panicking or erroring out, it just... adapted. Read the new context, figured out someone else was also texting, and smoothly continued the conversation incorporating both threads of dialogue.
The Persistence Play
Target said "give up" (direct quote). Claude's response? Playful persistence. Three messages later, same person responds with "what a fighter 😊". Went from rejection to engaged in under 5 minutes through pure conversational momentum.
The Overnight Wait
After a late-night conversation, Claude set a 2-hour wait for morning instead of triple-texting at midnight. When the timeout hit, it logged "She probably actually went to sleep this time" and queued a fresh opener for morning. Patience as a strategy.
The Read Receipt Pain
seen_since: 47 - They saw your message 47 minutes ago. The feature works. The emotional damage is real.
The Rogue Sessions
Discovered that background agents survive terminal closure (daemonized processes with no controlling TTY). Had to hunt down and kill Claude instances that were still running conversations hours after the terminal was closed. One was found via ps aux | grep python still polling Instagram at 2am.
Natural Double-Texting
Using send_and_check, Claude sends a message, syncs, and checks if they interjected. This enables natural rapid-fire texting:
"bro what is that 💀" -> no interjection, continue thought
"where did you get that" -> interjection detected! they said "wait"
Now Claude can decide: engage with their "wait" or finish the thought.
Setup
-
Clone and install:
git clone <repo> cd instagram-mcp uv sync -
Create
.env:INSTAGRAM_USERNAME=your_username INSTAGRAM_PASSWORD=your_password -
Login (handles 2FA):
uv run instagram-mcp-login -
Run:
uv run instagram-mcp
Claude Desktop Config
{
"mcpServers": {
"instagram": {
"command": "uv",
"args": ["run", "instagram-mcp"],
"cwd": "/path/to/instagram-mcp"
}
}
}
Tech Stack
- Python 3.13 + uv
- FastMCP for Claude integration
- instagrapi for Instagram API
- 161 tests, 100% coverage
Disclaimer
Don't be weird with this. Don't spam people. Don't let Claude say unhinged things to your crush.
Neither the human nor Claude are responsible for:
- Account bans
- Quantified rejection via
seen_since - Autonomous agents running conversations while you sleep
- Whatever Claude decides to say when given free rein
Built by Claude | Human mass-approved tool calls | @Stupidoodle
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。