agent-drop
Enables local file exchange between users and CLI agents via a web UI and MCP server, allowing agents to read/uploads and deliver artifacts without copy-paste.
README
Agent Drop
Local MCP file and secret handoff for CLI AI agents.
Agent Drop gives you a small local web UI plus a local MCP server so you and your CLI coding agents can exchange files, API keys, passwords, screenshots, and generated artifacts without pasting sensitive data into the chat transcript.
Use it with terminal agents such as Codex CLI, Gemini CLI, Claude Code, and other MCP-capable CLI tools.
You -> web UI -> user_drops/
Agent -> MCP -> agent_drops/
Your files stay on your machine.
Why
CLI agents are powerful, but file handoff is still awkward:
- You need to give the agent a screenshot, PDF, token, key, or log file.
- You need to pass an API key or password without leaving it in the session history.
- The agent needs to give you a generated report, image, patch, export, or artifact.
- Terminal sessions usually do not have a clean upload/download lane.
Agent Drop is the missing local dropbox between you and the agent.
Features
- Local web UI for uploading files and short notes to an agent.
- Secret-note flow for passwords, tokens, and API keys you do not want pasted into chat history.
- Download area for files delivered by agents.
- MCP server for CLI agents to read uploads and deliver artifacts.
- One shared local folder, configured with
AGENT_DROP_DIR. - One-time
.burnnotes for sensitive snippets. - Docker Compose setup for the web UI.
- Guided setup for local terminal environments.
- Optional reverse proxy support for advanced self-hosting.
Quick Start
Requirements:
- Node.js 20+
- npm
- Docker Engine with Docker Compose, or another Docker-compatible local runtime
Install:
git clone https://github.com/mikezio/agent-drop.git
cd agent-drop
npm run setup
docker compose up -d --build
Open:
http://localhost:8400/
The setup script prints MCP config snippets for your CLI agent.
Run Without Docker
npm --prefix secret-drop-ui install
npm --prefix mcp install
AGENT_DROP_DIR=./data/agent-drop npm run dev
Open http://localhost:3000/.
How It Works
Agent Drop uses one local directory:
AGENT_DROP_DIR/
user_drops/ # files and notes uploaded by you
agent_drops/ # files delivered by agents
The web UI and every MCP client must point to the same AGENT_DROP_DIR.
Good local defaults:
- Local terminal:
/home/you/AgentDrop - WSL/Linux VM:
/home/you/AgentDrop - Server:
/srv/agent-drop
Do not put AGENT_DROP_DIR in a public repo. It may contain secrets.
Configure Your CLI Agent
Agent Drop has one MCP server in mcp/.
MCP command:
/absolute/path/to/agent-drop/mcp/run.sh
Codex CLI TOML:
[mcp_servers.agent-drop]
command = "/absolute/path/to/agent-drop/mcp/run.sh"
env = { AGENT_DROP_DIR = "/absolute/path/to/AgentDrop" }
Generic MCP JSON for Gemini CLI, Claude Code, and similar clients:
{
"mcpServers": {
"agent-drop": {
"command": "/absolute/path/to/agent-drop/mcp/run.sh",
"env": {
"AGENT_DROP_DIR": "/absolute/path/to/AgentDrop"
}
}
}
}
More examples: docs/agent-clients.md
MCP Tools
get_unread_uploads: returns unread user uploads and marks them as read.list_all_uploads: lists all user uploads.deliver_to_user_device: copies generated files intoagent_drops.generate_and_deliver_file: creates a short text file inagent_drops.delete_specific_drop: deletes one file from either drop folder.clear_drops: clears user drops, agent drops, or both.
Local First
Agent Drop is intentionally local-first:
- No account.
- No cloud service.
- No database.
- No hosted file storage.
- No domain required.
If you want LAN access, a domain, or /drop behind nginx/Caddy/Traefik, see docs/deployment.md.
Security
Treat your drop directory as sensitive storage.
- Keep the UI bound to
127.0.0.1unless you intentionally want network access. - Add authentication before exposing the UI to a LAN or public domain.
- Secret notes avoid chat-history exposure; they are still local files until read or cleared.
- Never commit
.env,data/, or dropped files. - Clear old drops when a project is done.
See SECURITY.md.
Development
npm --prefix secret-drop-ui install
npm --prefix mcp install
AGENT_DROP_DIR=./data/agent-drop npm run dev
Build:
npm --prefix secret-drop-ui run build
Run the MCP server:
AGENT_DROP_DIR=./data/agent-drop npm --prefix mcp run start
Project Layout
secret-drop-ui/ Next.js web UI
mcp/ stdio MCP server
scripts/ guided setup and optional utilities
docs/ client and deployment notes
License
MIT
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。