Make AI Videos from Claude or ChatGPT with Biggsfield MCP

What the connector does

Biggsfield already has its own chat assistant, Biggs Agent, which plans shots, picks models, generates, edits and schedules recurring work while you approve. Many creators, though, spend their day inside Claude or ChatGPT. For them, Biggsfield can be connected as an MCP server. MCP (Model Context Protocol) is the standard that lets an AI assistant call outside tools. Once it is connected, the assistant you already use can read your request, choose a Biggsfield model, write a prompt in the format that model understands, start the generation and return the result.

The practical benefit is context. If you have spent an hour in Claude refining a script or a product launch plan, you don't need to copy everything into another tab. You can say "turn scene two into an 8-second clip" and the assistant already knows what scene two is.

Setting it up in Claude

In the Claude app (desktop or web), remote MCP servers are added as custom connectors. Open Settings, find the Connectors section, choose to add a custom connector, give it a name such as Biggsfield, and paste the MCP server URL that Biggsfield provides for your account. Save it and complete any sign-in step that appears so the connector acts on your Biggsfield account.

Before you start a chat, check that the connector is switched on for that conversation in the tools or connectors menu under the message box. Claude usually asks for permission the first time it calls a tool. You can approve each call individually, which is a sensible default for anything that spends credits.

Menu names in Claude change from time to time, and custom connector availability can depend on your Claude plan. If you can't find the option, look for Connectors or Integrations in Settings.

Setting it up in ChatGPT

ChatGPT supports custom MCP connectors through its connector settings. Open Settings, go to the Connectors (or Apps) area, and create a new connector using the same Biggsfield MCP server URL. Depending on your ChatGPT plan, you may first need to turn on developer mode before custom connectors appear.

After you add it, select the connector in a new chat so ChatGPT knows it may use it. As in Claude, keep confirmations on while you learn how the assistant interprets your requests. That way, a vague sentence won't turn into four 1080p renders you didn't intend.

The Claude Code command

If you work in the terminal with Claude Code, add Biggsfield as an HTTP MCP server with one command, replacing the placeholder with your Biggsfield MCP URL:

claude mcp add --transport http biggsfield <your-biggsfield-mcp-url>

Run /mcp inside Claude Code to confirm the server is connected and to finish authentication. This setup suits developers and content teams who script batches: for example, generating a product shot for every item in a CSV file, or creating launch visuals as part of a release checklist.

The seven tools, and what each one is for

The connector gives your assistant seven tools. You rarely need to call them by name, but knowing them helps you phrase requests and understand what the assistant is doing.

generate starts a video, image or audio generation on a chosen model. write_prompt turns your idea into a prompt written for a specific model. It is the same job Biggsfield's AI director does in the video studio, where each model gets the prompt it understands. run_app runs one of Biggsfield's one-tap apps, such as Relight, Virtual Try-On, Product Photo or a viral effect like Earth Zoom Out. get_generation checks the status of a job and returns the finished file.

list_models shows the available models with their durations, resolutions and options. list_apps lists the apps, which is useful because there are 85 of them. get_balance tells you how many credits you have left.

A good habit is to ask the assistant to call write_prompt first and show you the result before it calls generate. Reading the prompt costs nothing in render time, and fixing a wrong camera move in text is much cheaper than fixing it in a finished clip.

Example requests

Write requests the way you would brief a producer: what you want, which model if you care, duration, resolution and format. If you leave these out, the assistant will choose, and you may not agree with its choices.

Check my Biggsfield balance, then list the video models that support native sound and 1080p.

Use write_prompt for Veo 3.1 Fast: a barista slides a latte across a marble counter at golden hour, slow push-in, 50mm lens, ambient cafe chatter. Show me the prompt before generating a 4-second 720p test.

Generate the same idea on Kling 3.0 Pro at 5 seconds with sound, and write any dialogue in English.

Make a 30-second Seedance 2.5 sequence in three shots: Shot 1 wide street at dawn, Shot 2 medium on the runner tying her shoes, Shot 3 close-up as she says "Today" quietly.

Run the Relight app on this product photo with a studio look, then upscale it to 4K.

Create a Nano Banana 2.1 image at 1K of a ceramic mug on linen in soft window light. If I approve it, animate it with MiniMax Hailuo H3 Turbo for 5 seconds at 768P.

Choosing models from chat

Each model has its own prompting style, and the assistant follows it better when you name the model. Veo 3.1 and Veo 3.1 Fast want one continuous shot of up to 8 seconds, structured as subject and action, setting, camera, lighting, then audio, with dialogue written as Character says: "line" plus SFX: and Ambient: cues. Kling 3.0 Pro, Kling 3.0 4K and Kling O3 Pro respond to concrete physical action and strong camera language. Their native voice speaks only English or Chinese, so write dialogue in English. Seedance 2.5 handles several shots in one prompt (Shot 1, Shot 2) and refers to reference images as @Image1, @Image2.

For human performance without native audio, MiniMax Hailuo H3 Max describes the acting beat by beat. Wan 3.0 Prime follows long, detailed prompts and includes native audio at 1080p. For stills, Nano Banana 2.1 reads natural sentences rather than keyword lists, and GPT Image 2.5 is the choice when exact text or a logo has to appear, with every word in quotes.

When you are not sure which model to use, ask the assistant to run the same prompt on two models at a low resolution and compare. The Biggsfield video studio can also compare up to four models at once if you prefer to decide visually.

Costs: it all runs on your credits

Every generation made through the connector uses your Biggsfield credits, at the same prices as in the studio. Cheap drafts, for reference: a 5-second MiniMax Hailuo H3 Turbo clip at 768P is 34 credits (about $0.33 on Creator), and a 4-second Veo 3.1 Fast clip at 720p with sound is 100 credits (about $0.96). A 5-second Wan 3.0 Prime clip at 720p with sound is 117 credits, and a 5-second Kling 3.0 Pro clip with sound is 140 credits.

Final-quality renders cost more. An 8-second Seedance 2.5 clip at 720p with sound is 631 credits, at 1080p it is 1552 credits, and a 15-second Kling 3.0 4K clip with sound is 1050 credits. Images are much cheaper: Nano Banana 2.1 starts at 8 credits for a 1K image, and GPT Image 2.5 is 25 credits.

Because a chat assistant can start several jobs from one sentence, set rules at the start of the conversation. For example: "Always tell me the model, duration, resolution and credit cost before calling generate, and never run more than one job without asking." Calling get_balance before and after a session also gives you a clear record of what it cost.

A workflow that works

Plan in conversation, draft cheap, finish expensive. Use the assistant to write the script and shot list. Have write_prompt draft model-specific prompts, then test motion and framing at the lowest resolution the model offers. Use get_generation to review each result. Only move to 1080p or 4K once the shot is right.

For finishing, you can still go back to the Biggsfield studio. Biggs Cut gives you a multi-track timeline with auto subtitles and cloud rendering, and Biggs Director lets you lock camera, lens and cast across a storyboard. The connector is the fast way in. The studio is where you polish the cut.

Questions

More guides