Pandafile Pro · AI assistants

Send files from Claude, ChatGPT, Cursor and other AI assistants

Ask your AI assistant to send a file and get a download link back: "Send the renders folder to client@example.com and keep it for 60 days." The official Pandafile MCP server gives Claude, Cursor, VS Code, Windsurf and other assistants the tools to upload big files and whole folders, email recipients and keep track of who downloaded what.

AI assistants are good at deciding what to send and to whom, but they cannot attach a 40 GB video to a chat message. Pandafile does the heavy lifting: the files are uploaded from your own computer with resumable, chunked uploads, and the assistant only receives the link, the transfer id, the sizes and the expiry. It works through the open Model Context Protocol (MCP), the standard way for assistants to use outside tools.

Which AI assistants can send files with Pandafile?

The Pandafile MCP server is a local server: your assistant starts it on your computer with npx -y @pandafile/mcp and talks to it over stdio. Every client that can start local MCP servers can use it.

AssistantWorks?How
Claude DesktopYesAdd the server to claude_desktop_config.json
Claude CodeYesclaude mcp add
CursorYes.cursor/mcp.json
VS Code (GitHub Copilot agent mode)Yes.vscode/mcp.json
WindsurfYes~/.codeium/windsurf/mcp_config.json
OpenAI Codex CLIYescodex mcp add
ChatGPT (web and apps)Not yetChatGPT connectors only reach remote MCP servers over HTTPS; this server runs locally

Using ChatGPT? Until there is a remote Pandafile server, send from pandafile.co in the browser or with the pandafile command line – or let ChatGPT write a script for you with the Python SDK or the Node.js SDK.

Set it up in three steps

1. Get an API key. The API is included with Pandafile Pro and Team (see pricing). Create a key in Dashboard → Domain & API. While you try things out, use a test key (pf_test_…): uploads are real, but nobody is emailed and test transfers expire after 24 hours.

2. Add the server to your assistant. You need Node.js 18 or newer (it includes npx). Replace pf_live_YOUR_KEY with your key.

Claude Desktop

// Settings → Developer → Edit Config (claude_desktop_config.json), then restart Claude
{
  "mcpServers": {
    "pandafile": {
      "command": "npx",
      "args": ["-y", "@pandafile/mcp"],
      "env": { "PANDAFILE_API_KEY": "pf_live_YOUR_KEY" }
    }
  }
}

Claude Code

claude mcp add pandafile --env PANDAFILE_API_KEY=pf_live_YOUR_KEY -- npx -y @pandafile/mcp
# add --scope user to have it in every project

Cursor

// .cursor/mcp.json in your project, or ~/.cursor/mcp.json for every project
{
  "mcpServers": {
    "pandafile": {
      "command": "npx",
      "args": ["-y", "@pandafile/mcp"],
      "env": { "PANDAFILE_API_KEY": "pf_live_YOUR_KEY" }
    }
  }
}

VS Code

// .vscode/mcp.json – VS Code asks for the key once and stores it securely
{
  "inputs": [
    { "type": "promptString", "id": "pandafile-api-key", "description": "Pandafile API key", "password": true }
  ],
  "servers": {
    "pandafile": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@pandafile/mcp"],
      "env": { "PANDAFILE_API_KEY": "${input:pandafile-api-key}" }
    }
  }
}

Windsurf

// ~/.codeium/windsurf/mcp_config.json
{
  "mcpServers": {
    "pandafile": {
      "command": "npx",
      "args": ["-y", "@pandafile/mcp"],
      "env": { "PANDAFILE_API_KEY": "pf_live_YOUR_KEY" }
    }
  }
}

Codex CLI and others

# Any client that starts local (stdio) MCP servers:
#   command  npx -y @pandafile/mcp
#   env      PANDAFILE_API_KEY=pf_live_YOUR_KEY
codex mcp add pandafile --env PANDAFILE_API_KEY=pf_live_YOUR_KEY -- npx -y @pandafile/mcp

3. Ask. Restart the assistant if it asks you to, then talk to it in plain words. It picks the right tool, asks for anything that is missing and gives you the link.

Things you can ask

  • "Send ~/Desktop/final-cut.mov to client@example.com with Pandafile and give me the link."
  • "Upload the renders/ folder, title it Final renders, put the password blue-river-42 on it and keep it for 60 days."
  • "Has the client downloaded yesterday's transfer? If not, send them a reminder."
  • "Make a file request called Wedding photos so the couple can upload to me, and give me the upload link."
  • "Turn these three public https links into one Pandafile transfer." (Pandafile downloads them itself.)
  • "Which of my transfers expire this week?"

Behind those requests are thirteen tools – send_files, send_from_urls, list_transfers, get_transfer, add_recipients, remind_recipient, update_transfer, delete_transfer, create_file_request and more. The full list with what each does is on the MCP server page.

How it works, and what the AI sees

When you ask for a file to be sent, the assistant calls send_files with the paths. The MCP server, running on your computer, reads exactly those files, creates a transfer and uploads them straight to Pandafile's servers in Helsinki, Finland (EU), in chunks that resume after a dropped connection. Folders keep their structure. When the upload is complete, the server returns the link, id, expiry and sizes to the assistant. The contents of your files are never handed to the model or to the AI company – only the result of the tool call is.

The server is built on the official JavaScript SDK, so every request carries an idempotency key and is retried safely, and it sends progress notifications while a big upload runs. API errors come back as readable messages with a request id instead of crashing the tool.

Stay in control

  • Confirmation before changes. Every tool is annotated as read-only, writing or destructive. delete_transfer is destructive, so clients such as Claude and Cursor ask you first.
  • Narrow keys. API keys have a scope – read, write or full. An assistant that only needs to report on downloads can get a read key.
  • No telemetry. The server reads only the paths it is given, and your API key goes only to the Pandafile API; it is never logged or shown in tool results.
  • Test mode. With a pf_test_ key nothing is emailed and everything disappears after 24 hours.

Limits and price

Transfers sent by an assistant follow the same rules as the website. Pro sends up to 1000 GB per transfer with 1500 GB of storage, keeps files for 30 days (extendable up to 90) and includes the API, file requests, your own branding and your own download domain. Team gives up to 5 people everything in Pro on one invoice. Prices are on the pricing page.

Building something yourself rather than chatting? The API reference documents every endpoint, the SDKs and CLI page covers JavaScript, Python and the command line, and the file transfer API overview explains what you can automate.

Frequently asked questions

Can ChatGPT send files with Pandafile?

Not yet. ChatGPT connects only to remote MCP servers over HTTPS, and the Pandafile MCP server runs locally on your computer (stdio). Use Claude Desktop, Claude Code, Cursor, VS Code, Windsurf or the OpenAI Codex CLI, or send from pandafile.co or the pandafile command line.

Do my files go through the AI company?

No. The MCP server runs on your computer, reads the files from the paths the assistant passes to it and uploads them straight to Pandafile in Finland (EU). The model only sees the tool result: the link, the transfer id, sizes and the expiry.

Which plan do I need?

The API, and so the MCP server, is included with Pandafile Pro (from $5.99 / €5.99 a month) and Team. A test key (pf_test_…) lets you try it: real uploads, but nobody is emailed and they expire after 24 hours.

How big can the files be?

As big as your plan allows: up to 1000 GB per transfer on Pro and Team. Uploads are chunked and resumable, so a dropped connection continues where it stopped.

Can the assistant delete my transfers without asking?

Every tool carries MCP annotations (read-only, destructive, idempotent). delete_transfer is marked destructive, so clients such as Claude and Cursor ask you before it runs. You can also give the assistant a read or write key instead of a full key.