The problem
You ask ChatGPT for a chart, then ask Claude Code to put that chart in the project's docs. Today that means downloading the image, finding it again, and uploading it somewhere Claude can reach. The same happens with logos, screenshots, exported reports and anything else one tool makes and another needs.
How it works with drop.
Each tool connects to drop. once, through the MCP server at https://drop.media/mcp, the drop command-line tool or the REST API. After that:
- When a tool makes a file, it uploads it and gets a link back.
- When you mention a file ("the Q3 chart", "our logo"), the tool searches your library with
list_filesbefore making a new one. - Anything you upload yourself on the website is in the same library, so your agents can find that too.
In practice the conversation goes like this:
# in Claude Code
You: Put the Q3 revenue chart in docs/board-update.md
Claude: list_files "q3 revenue" → 1 file (uploaded from ChatGPT on Tuesday)
Added 
Connecting your tools
ChatGPT, Claude and other apps that support MCP connectors can add https://drop.media/mcp and sign in once. Tools that take an API key instead, such as Claude Code and Cursor, use a key from the Developers page:
claude mcp add --transport http drop https://drop.media/mcp \ --header "Authorization: Bearer <api key>"
For anything with a shell, the CLI works too: npm install -g drop-media, then drop ls to search and drop file.png to upload.
Keeping each tool in its lane
Each API key has its own abilities: files:read, files:write, files:delete and images:generate. An agent only sees the MCP tools its key allows, so you can give a research assistant read-only access and keep delete rights for yourself. Revoke a key and that tool loses access straight away; the others carry on.