Log inStart creating

Making videos from Claude and other agents with an MCP server

What an MCP server is, how AI agents like Claude use one to make videos, how to set it up safely and a sample request flow. Concepts only; see the MCP docs.

, 4 min read

The short answer

An MCP server lets an AI assistant such as Claude use a service's tools on your behalf. With ClickPop's MCP server connected, you can ask an agent in plain language to plan and make a video, and it calls ClickPop to do the work.

This guide explains the idea and a safe way to work with it. For setup steps and the tools available, follow ClickPop's MCP documentation at /docs/mcp.

What MCP is

MCP stands for Model Context Protocol, an open standard for connecting AI assistants to outside tools and data. A service publishes an MCP server that describes what it can do. An AI client that supports MCP connects to it, sees the available tools and can call them when a conversation needs them.

For you, the result is that the agent is no longer limited to writing text about a video. It can start the video, check on it and bring the result back into the conversation.

The difference from an API is who decides the steps. With an API, your code calls each step in a fixed order. With MCP, the agent reads your request, chooses which tools to call and in what order, and explains what it did. That makes MCP good for one-off and exploratory work, and an API better for fixed, high-volume pipelines.

What an agent can do with video tools

The exact tools depend on the server, so check /docs/mcp. In general, an agent with video tools can help with work like this:

  • Turn a rough idea into a brief, a script and a shot list, then generate the shots.
  • Make several variants of an ad hook for testing.
  • Produce a short explainer from notes or a document you share in the chat.
  • Check the cost of a plan before spending credits.
  • Keep a consistent character across several videos by reusing saved references.

The agent is good at planning, writing and repetitive steps. You stay in charge of taste, approval and anything that goes public.

Setting it up

Setup usually follows the same pattern, though the details live in the docs.

  • Read /docs/mcp for the server address and the supported clients.
  • Add the server in your AI client's connector or MCP settings.
  • Sign in or authorize access to your ClickPop account as the docs describe.
  • Start a new conversation and ask the agent what video tools it now has.
  • Run a small test, such as one short clip, before giving it larger jobs.

Treat the connection like any account access. Only connect clients you trust, and disconnect ones you no longer use.

A sample request flow

Here is how a session might go. The wording is yours; the agent picks the tools.

  • You: "Make a 15-second vertical video for our ceramic mug launch. Warm morning kitchen, our host character, voiceover in English, captions on. Show me the plan and cost first." Try it
  • Agent: writes a four-shot plan with a hook, voiceover lines and camera moves, and reports the estimated credits.
  • You: "Swap shot 3 for a close-up of the glaze. Then go ahead."
  • Agent: generates the shots, assembles the cut and returns a link.
  • You: "Make two more hooks for testing, same shots."
  • Agent: produces two variants and lists what changed in each.

Asking for the plan and cost first is the important habit. It keeps you in the loop before credits are spent and gives you a chance to fix the idea while it is still cheap.

Write better requests

Agents work best with the same information a human producer needs. A good request names the goal, audience, platform, length, aspect ratio, character, look, voice and what to avoid.

For example: "Goal: drive clicks to the product page. Audience: home cooks, 25-45. Platform: Instagram Reels, 9:16, 20 seconds. Character: Mara from our saved characters. Look: warm film grade, soft window light. Voice: calm English. Avoid: text in the scene, other brands, any health claims."

Try this prompt in Create Video

Save that as a template and paste it at the start of each request. The agent will follow it more consistently than it will follow a one-line ask.

You can also share source material. Paste product notes, a blog post or a list of customer questions into the chat and ask the agent to turn it into a script first. Review the script, fix anything inaccurate, and only then ask it to generate. Catching an error in a script costs nothing; catching it after rendering costs credits and time.

Safety and control

An agent acts with your account, so set boundaries.

  • Ask for plans and costs before generation, every time.
  • Set a credit budget in your request: "Do not spend more than X credits without asking."
  • Review every video before it is posted. The agent should not publish on its own unless you have built that step on purpose with checks.
  • Keep the content rules: no real people's likeness without permission, no third-party brands or logos.
  • Label realistic AI content where platforms ask, as of October 2026.

Mistakes to avoid

  • Assuming the agent knows your brand. Give it a written brief every time.
  • Letting an agent run a large batch before testing a single clip.
  • Guessing tool names or options instead of checking /docs/mcp.
  • Connecting the server to a shared or untrusted client.

Questions

What is an MCP server?

It is a connector that follows the Model Context Protocol, letting AI assistants call a service's tools. ClickPop's MCP server lets agents like Claude make videos.

Can Claude make videos?

With ClickPop's MCP server connected to a client that supports MCP, Claude can plan and generate videos through ClickPop. See /docs/mcp for setup.

Is MCP different from the API?

The API is for code you write. The MCP server is for AI agents that decide which tools to call during a conversation. Both use your ClickPop account.

Making videos from Claude and other agents with an MCP server | Click Pop