What is an MCP image server, and why use one?

Start with the protocol

MCP, the Model Context Protocol, is a standard way for an AI agent to call outside tools. The agent reads a list of tools a server offers, then calls them with structured arguments and gets structured results back. That is the whole idea.

An MCP image server is just an MCP server whose tools happen to generate pictures. A video server does the same for clips. Figgshield is both, reachable at https://api.figgshield.ai/mcp. Your agent connects once and sees the generate tools next to everything else it can do.

So there is nothing magic here. If you have written a tool call before, you already understand the shape of it. The server does the heavy model work and hands back a link to the finished file.

What the agent actually calls

Once Figgshield is connected, your agent gains a tool to generate an image and a tool to generate a video. It fills in a prompt, a size, and a few options, then sends the call. You never touch an API key in your code or manage a model endpoint yourself.

The image tool runs Nano Banana 2 from Google. A render at 1K or 2K costs 8 credits, and 4K costs 16. The video tools run Kling 3, with a faster Kling 3 Turbo tier when you want speed over polish. A five second clip is 35 credits. One balance covers both, so you are not juggling separate wallets for stills and motion.

Because the agent sees the result as structured data, it can keep going on its own. It can generate an image, read back the file id, and start a video from that same image without you copying anything between windows.

The trouble with a web tool

A browser tool wants your full attention. You leave your editor, log in, type a prompt, wait, download the file, rename it, and drag it into your project. Every image is a small errand, and every errand breaks your focus.

None of that is scriptable where you actually work. Your agent cannot open a tab for you, and it cannot fold the result into the next step of a task. As far as automation goes, the web tool is a dead end.

A server closes that gap. The agent asks for the media it needs during a task and uses the result right away. You stay in Claude Code, Cursor, or wherever you were, and the picture shows up as part of the conversation.

Pricing you can actually reason about

Figgshield charges at cost. We pay the model providers and pass that price straight through, with no markup on Nano Banana 2 or Kling 3. The credit numbers above are the real cost of the render, not a marked up sticker price.

So how is that a business? The only money we make is on credits people buy and never spend. That lines our incentive up with yours, because we would rather you use what you paid for than feel shortchanged.

You can start on the $5 seven day Pro trial, which includes 150 credits. There is no free tier. If it is not for you, cancel before day 7 and you pay nothing more. The credits page shows how far a balance goes.

Connecting it to your editor

Figgshield speaks plain MCP, so it works with the clients most developers already run. That covers Claude on the desktop and in Claude Code, Cursor, Windsurf, and ChatGPT in Developer Mode on a paid plan. Any MCP capable agent can use it.

1. Open your client's MCP settings and add a new server.
2. Point it at https://api.figgshield.ai/mcp and sign in.
3. Ask your agent to generate a test image to confirm the tools are live.

Where your files live

Every image or clip you generate sits on a private link for 7 days. After that it is deleted. The window is long enough to pull the file into your project and short enough that nothing lingers on our side.

If you send a reference image to guide a render, it is used in memory for that request and never stored. Once the response is sent, it does not sit on a disk on our side.

For a lot of teams that is the deciding factor. You get the output you need without a growing pile of your material living on someone else's server.