How Do You Generate AI Images and Videos From Claude or ChatGPT With the VdoBloom MCP Server?

You can generate a VdoBloom image or video from inside Claude, ChatGPT or Cursor by connecting one address, https://vdobloom.com/mcp, and letting the assistant call seven tools against your own VdoBloom credit balance. There is nothing to install: the connector signs you in, the tools list models and prices before they spend anything, and an image on the default model costs 6 credits.

Updated October 2026.

What an MCP server lets your AI assistant do

Model Context Protocol is a standard way for an AI assistant to reach tools outside its own chat window. Without it, asking Claude to make a video gets you a description of a video. With a VdoBloom MCP connection, the assistant can call the platform directly: list the models, read the balance, send a prompt, check the job, and hand back a link to the finished file.

The key detail is whose account pays. The connector is authorised by you and charges your VdoBloom balance at the same prices as the web app. The assistant sees prices through the tool list, so it can tell you what a clip costs before committing the run. It is useful for drafting creative in chat, running a few variations without switching tabs, and reaching for an image as one step in a longer task. The developer page documents the server alongside the REST API.

Connect VdoBloom in Claude or ChatGPT as a custom connector (sign in, no key to paste)

Both apps take the same input: the server URL, https://vdobloom.com/mcp, added as a custom connector. Paste it, start the connection, and a VdoBloom sign-in page opens in a browser.

The sign-in uses OAuth 2.1 with PKCE and dynamic client registration. In plain terms, the client registers itself, the browser proves the sign-in with a one-time code, and no client secret is created or stored. A consent screen names the client and the redirect host, and you approve it before any token is issued. The connection then appears in VdoBloom under Settings, Developer as a key named after the client, which is also the switch that disconnects it later.

Command-line clients: claude mcp add with a vdb_ key

For a terminal client, create a key first. Open Settings, Developer in the dashboard and create one. The key starts with vdb_, it is shown once, and only a hash is stored, so copy it at that moment. The same page can revoke it.

Then add the server in one command:

claude mcp add --transport http vdobloom https://vdobloom.com/mcp --header "Authorization: Bearer vdb_your_key"

The header is the API's bearer format, so an existing key works here. A key can generate and read jobs but cannot create or revoke other keys, which keeps a leaked client from widening its own access.

The seven tools and what each one does

The server exposes a small, fixed set of seven tools.

Tool	What it does	Main parameters
list_models	Lists image, image-edit and video models with credit prices	type, model (for a full price table)
get_balance	Returns your remaining credits	none
generate_image	Text to image	prompt, model, aspect_ratio, resolution
edit_image	Edits or combines 1 to 10 images	prompt, image_urls, model, aspect_ratio, resolution
generate_video	Text to video, or image to video with a first frame	prompt, model, image_url, duration, resolution, aspect_ratio
get_status	Returns processing, completed or failed for a job	job_id
list_creations	Lists your recent images and videos, newest first	type, limit, before

Defaults are sensible. An image call goes to Nano Banana unless you name another model, and a video call goes to Seedance 2 Fast. Both defaults can be overridden per call, and list_models reports the cheapest option for every model it returns. Images passed to edit_image or used as a first frame must be public https links, because the server fetches them once and keeps its own copy.

A first session: product image, 5-second clip, three variations

A realistic first session with a product photo looks like this. Start in plain language and let the assistant choose the tools.

Check the balance and the price. Ask what the account has and what a 5-second clip costs on Seedance 2 Fast. The assistant calls get_balance and list_models, then reports the numbers. A 5-second 480p run is 24 credits and a 5-second 720p run is 50 credits, so the trade is clear before anything renders.
Make the product image. Ask for a studio shot of your product on a plain surface with directional light and space above for a headline. The assistant calls generate_image with Nano Banana at 6 credits, and usually returns the finished image in the same turn.
Produce three variations. Ask for the same product on three backgrounds: warm oak, pale grey seamless, and a dark slate slab. With edit_image, each variation is a separate instruction against the original image, at 7 credits a call.
Animate the winner. Ask for a 5-second 9:16 clip that starts from the chosen still: a slow push in while a soft light sweep crosses the surface. Pass the image link as image_url and the assistant calls generate_video.
Check the job. Video returns a job immediately. Ask for the status, and the assistant calls get_status until it reports completed, then gives you the video link.

That is four images and one clip for 51 credits: 6 for the base image, 21 for three edits, and 24 for the video at 480p. If you would rather build the prompt before you start, how to write a better AI image prompt covers the parts that change the output. Running many renders in one session is covered in AI video batch generation.

Credits, refunds and limits (same prices as the app, failed jobs refunded, revoke in Settings)

The connector spends the same credits as the web app, at the same prices, with no separate developer plan. Credits you buy do not expire, so a balance funded for one project is still there for the next. See AI video credits explained and pricing. Every generation shows its cost, and the assistant is instructed to check prices and balance before a large request.

A failed generation is refunded automatically. A blocked prompt fails the job rather than charging for it, because prompts go through the same content rules as the app. If the balance runs out mid-session, generation calls fail with an insufficient-credits error and nothing is deducted. Connections live in Settings, Developer: each app is listed as its own key, and revoking it disconnects that client.

What MCP does not do (effects, lip-sync and ad tools stay in the dashboard)

The seven tools cover image generation, image editing, video generation, job status, balance, model prices and recent creations. They do not cover the rest of the platform. Effects such as dance or fashion presets, lip-sync and avatar work, UGC and ad tools, and timeline editing all stay in the app.

A chat window is good at describing a shot and running a few versions of it. A timeline is good at assembling them. When a job needs a talking presenter, an ad wizard or a cut on a timeline, move to the dashboard: text to video for model picking, image generation for stills, and the editor for assembly. If an assistant says it can run a lip-sync job through this connection, it is describing something the server does not expose.

Frequently asked questions

Do I need an API key to use VdoBloom in Claude or ChatGPT?

No. Add https://vdobloom.com/mcp as a custom connector and sign in through the browser. The OAuth flow issues access without a key to paste. Command-line clients can use a vdb_ key with an Authorization header instead.

Which models does the connector use by default?

Images default to Nano Banana and videos default to Seedance 2 Fast. You can name a different model on any call, and list_models shows the price of each before you commit.

How much will a first session cost?

A base product image is 6 credits, an edit is 7, and a 5-second 480p Seedance 2 Fast clip is 24. The session above comes to 51 credits, and the assistant can quote every price from list_models first.

Can the assistant edit an image I already generated?

Yes. edit_image takes a prompt and one to ten public image URLs, so it can take the last image, change the background, or combine several references in one instruction.

What happens when a video job fails?

get_status reports failed with an error, and the generation is refunded. Check the prompt against the platform content rules first, then retry with the same idea phrased more plainly.

Open image generation or text to video to see the same models and prices in the app, and read the developer docs for the full setup. New accounts start with 10 free credits and no card, so you can connect an assistant and run one image before deciding whether the workflow suits you. There is no subscription needed, and one-time credits never expire.