How to Use MCP: Connect Claude, ChatGPT, and AI Agents to YourImageShare

By Product Team · Jul 28, 2026 · 5 min read · Last updated Aug 3, 2026

Most people uploading a screenshot or a photo still do it the old way: open a browser tab, drag a file in, copy a link. But a growing number of tools now talk to AI agents directly instead of through a browser, using a standard called MCP, the Model Context Protocol. If you use Claude, or increasingly ChatGPT, you can already give it the ability to upload, list, and delete files on YourImageShare as part of a conversation, without ever opening this site yourself.

What MCP actually is

MCP is an open protocol, originally introduced by Anthropic, that lets an AI assistant call external tools in a standardized way. Instead of every company inventing its own plugin format, an MCP server describes a small set of tools it exposes, upload a file, list your files, delete a file, in our case, and any MCP-compatible client can discover and call them the same way. It's the same idea as a REST API, except the client is an AI agent instead of your own code, and the "documentation" the agent reads is the tool schema itself.

Why this is worth knowing about now

Support for MCP is still uneven across the major assistants, which is exactly why it's worth getting ahead of. Claude Desktop and Claude Code have supported MCP natively since it launched. ChatGPT added MCP support and, as of early 2026, fully supports the related "MCP Apps" standard for embedded tool UIs, so a well-built MCP server works there too. Gemini is further behind, it currently understands MCP-style tool schemas inside its own function-calling API rather than acting as a full MCP client that connects to arbitrary servers. In practice, that means Claude and ChatGPT users can use an MCP-connected upload tool today, and that gap is unlikely to stay open for long once more assistants catch up.

Setting up the YourImageShare MCP server

The official YourImageShare MCP server needs no install step, it runs directly with npx:

npx yourimageshare-mcp

To use it with Claude Desktop or Claude Code, add it to your MCP configuration with your API key (found under the API tab in My account):

{ "mcpServers": { "yourimageshare": { "command": "npx", "args": ["yourimageshare-mcp"], "env": { "YIS_API_KEY": "YOUR_API_KEY" }}}}

Restart the client and it will pick up three new tools: upload a file, list your uploads, and delete an upload by id, the same three endpoints documented in full on the API reference page. For ChatGPT, the setup lives under its connector or MCP server settings rather than a local config file, but it points at the same server and the same API key.

What this actually looks like in a conversation

Once it's connected, uploading stops being a separate step. You can drop an image into a chat and ask Claude to "upload this and give me a direct link," and it calls the tool itself and hands back the URL. You can ask it to "list what I've uploaded this week" or "delete the last three test uploads," and it does that through the same MCP tools rather than you opening My account and clicking through the list by hand. For anyone building an agent that generates images, screenshots dashboards, or produces reports with visuals attached, the agent can host the output itself and include a real link in whatever it hands back, without a separate upload step bolted on afterward.

Where this fits next to the REST API

MCP isn't a replacement for the plain HTTP API, it's a different front door onto the same three endpoints. If you're writing a script, a cron job, or a backend service, the REST API and the official JavaScript and Python libraries covered in a previous post are still the right tool. MCP matters specifically when the thing making the upload decision is an AI agent acting inside a chat client, not a script you wrote and control line by line.

A note on keys

The same rule applies here as everywhere else the API key is used: treat it like a password. Anyone with access to your MCP config can upload, list, and delete on your account through the assistant. If a key ends up somewhere it shouldn't, regenerate it from the API tab in My account, the old one stops working immediately.

Where to go from here

Full setup details, along with the JavaScript, Python, and MCP client libraries, live on the API documentation page. It's a five-minute setup, and it's one of the few upload tools right now that any MCP-compatible AI agent can use directly, not just Claude.