Every other AI

If it speaks MCP, it works. ChatGPT, Claude and Gemini get their own pages because they are what most people arrive with. Everything else (every other client, and every open-weight model) connects the same way, and this page is that way.

Prism keeps no allow-list. It exposes a standard MCP endpoint over HTTP and never inspects which model is answering.

What "bring your own AI" means

Prism is built on the Model Context Protocol (MCP), an open standard for connecting AI clients to tools, so any client that speaks MCP can connect to it, that's what an MCP server for After Effects is.

You keep your existing subscription with your AI provider and your existing workflow. Prism adds one capability to that agent: the ability to read and write your live After Effects project, with the server holding all the logic.

Why AI actions have no cap

Most tools that bundle their own AI have to pay a model provider for every message, so they meter your usage to cover that cost. Prism does not bundle a model (you supply your own), so there is no provider bill for Prism to pass along. That is the whole reason AI actions are unlimited: there is nothing to recover.

The only metered capability is the one where Prism itself does real work on your behalf: the AI balance that pays for the generative tools, in dollars. Everything the agent reads and writes directly in your comp is unlimited on Pro, which gives unlimited AI actions; Free includes 200 AI actions per month, and every account starts with $1.00 of AI balance at signup. For caps, fail-open behavior, and the per-tool prices, see usage and limits.

Clients that use this page

These all take the same URL. The only difference is where you paste it.

Open-source and self-hosted models

Prism does not care which model is on the other end. Llama, Mistral, DeepSeek, Qwen, Kimi, or a model you run yourself on your own hardware: each drives After Effects exactly as a frontier model does, provided the client wrapped around it can add a remote MCP server and call tools.

That is the entire requirement. Model quality changes how well the agent plans, not whether Prism answers it.

The generic setup

{
  "mcpServers": {
    "prism": {
      "url": "YOUR_PRISM_MCP_URL"
    }
  }
}

Some clients name that key serverUrl or httpUrl. The value is the same either way: the URL the panel gave you.

What your client must support

A client that only supports local stdio MCP servers cannot reach Prism today.

Common questions

Does the model matter?

No. Prism is model agnostic. A stronger model plans better and asks for better things, but there is no model Prism requires and none it refuses.

Can I use an open-source model I host myself?

Yes. If your client can add a remote MCP server and call tools, a self-hosted model works exactly like a hosted one. Nothing about Prism changes.

Can I connect more than one client?

Yes. The URL is not exclusive, so add it to every client you use and switch whenever you like.

My client supports MCP but will not connect. What now?

Check the panel is open in After Effects and a composition is active. Those two account for nearly every failed connection. Then confirm your client supports remote MCP servers rather than only local ones. The rest is in troubleshooting.

Related

• Your AI client: the whole picture
• ChatGPT · Claude · Gemini
• What an MCP server for After Effects is: the architecture