How to Use Seedream 5.0 Pro in Claude Code and Cursor?

Seedream 5.0 Pro runs in Claude Code and Cursor through the Makernice MCP server: the agent writes the prompt, prices the image with estimate_cost and calls generate once you approve. A text-to-image result at the default 1k is $0.0351 per image at price list v15 (November 12, 2026).

Connecting the agent to Makernice

In a terminal, run the claude mcp add --transport http --scope user makernice command shown on the Claude Code instructions; the page includes the address to append. Inside Claude Code, /mcp lists makernice, and selecting it opens the browser sign-in.

In Cursor, the Cursor instructions give you an install button and the ~/.cursor/mcp.json entry. Then use the sign-in control next to makernice in Cursor's MCP settings. With either editor, a quick get_balance call confirms the connection and shows the per-job limit it will enforce.

Where can I use Seedream 5.0 Pro: an app, an API or another platform?

Seedream 5.0 Pro is available on Makernice in three places at the same per-image price: the REST API, the MCP server used by Claude Code, Cursor, Claude and ChatGPT, and the Create form in the Makernice console for running it in a browser. Other API platforms sell Seedream 5.0 Pro too, and the third-party price table on the Seedream 5.0 Pro model page lists the ones that post a per-image price for the same tier, with the date each price was checked. ByteDance's own apps are outside this guide, and we haven't checked them.

Can I make Seedream 5.0 Pro images in Claude or ChatGPT instead?

Yes, and only the connection differs. Once Makernice is added, the chat app calls estimate_cost, waits for your go-ahead, sends generate with that figure as max_cost_usd and reads the result from check_job, the same three calls Claude Code and Cursor make for Seedream 5.0 Pro. The estimate is not a reserved price: generate quotes again and stops if the new quote is over your limit.

• Claude: Customize → Connectors → + Add → Add custom connector, paste the MCP URL, then Connect and sign in. On Team and Enterprise plans an owner adds it in Organization settings first. Claude steps.
• ChatGPT on the web: Plugins → + → Add custom MCP server, Authentication set to OAuth and the ChatGPT-only URL, then type @ in a new chat and pick Makernice. Your plan or workspace must allow custom MCP servers, and that version has no get_topup_link tool, so top up on the site. ChatGPT steps.

One Seedream 5.0 Pro image, priced and then made

You: I need a portrait-format product shot: perfume bottle on wet slate, orchid in the background. Quote it before you make it.

Claude Code sends:

{
  "name": "estimate_cost",
  "arguments": {
    "input": {
      "model": "seedream-5.0-pro",
      "mode": "t2i",
      "prompt": "Editorial still life of a glass perfume bottle on wet black slate, a single orchid stem behind it, soft rim light, high detail",
      "resolution": "1k",
      "aspect_ratio": "3:4"
    }
  }
}

Claude Code: One 1k image from Seedream 5.0 Pro comes to $0.0351 on price list v15. I haven't created anything. Go ahead?

You: Please do.

{
  "name": "generate",
  "arguments": {
    "input": {
      "model": "seedream-5.0-pro",
      "mode": "t2i",
      "prompt": "Editorial still life of a glass perfume bottle on wet black slate, a single orchid stem behind it, soft rim light, high detail",
      "resolution": "1k",
      "aspect_ratio": "3:4"
    },
    "max_cost_usd": "0.0351",
    "request_id": "perfume-slate-portrait-1k"
  }
}

The max_cost_usd value is the estimate you accepted. When generate runs, Makernice quotes the same input again; a quote above max_cost_usd or above the connection's per-job limit stops the call before any job exists. Otherwise the job is created at that quote and a job_id comes back. Reusing the request_id with unchanged input after a dropped reply returns this job, so the image is never paid for twice.

{
  "name": "check_job",
  "arguments": {
    "job_id": "gen_...",
    "wait_s": 20
  }
}

An image may be ready within a single check_job wait; if not, the agent calls again. The result is at outputs[0].url as a signed link that lasts 24 hours; a new check_job call returns a fresh link and console_url points to the run in your console. Prefer push notifications? Your account webhook covers MCP jobs as well.

How do I edit an image with Seedream 5.0 Pro, and are layers or inpainting supported?

Seedream 5.0 Pro edits an image in image-to-image mode: set mode to i2i, send the source picture as a public https link in reference_images, and write the change in prompt, such as swapping the label on a bottle or moving a product onto a new surface. The request has no mask field, so painting over an area to inpaint it is not supported; name the area in words instead. Layered output is not supported either, and each Seedream 5.0 Pro job returns one flattened image. For a cut-out on a transparent background, Seedream 5.5 takes a background field set to transparent.

How do I use multiple reference images with Seedream 5.0 Pro, and how are they billed?

Seedream 5.0 Pro accepts 1 to 10 public https links in reference_images for one edit, and the base edit price already covers the first of them. The price has two parts. The base edit price, $0.0351 at 1k or $0.0702 at 2k, already covers the first reference image. Every reference after the first adds $0.00234. In list_models output those two figures appear as unit_price_usd and per_reference_price_usd, with free_references saying how many images the base price includes.

Here the agent prices a 2k composite built from three references, which means two extra reference charges on top of the base. The estimate returns the combined total:

{
  "name": "estimate_cost",
  "arguments": {
    "input": {
      "model": "seedream-5.0-pro",
      "mode": "i2i",
      "prompt": "Place the perfume bottle from image 1 on the marble shelf from image 2, lit like the reference in image 3",
      "reference_images": [
        "https://example.com/assets/bottle-cutout.png",
        "https://example.com/assets/marble-shelf.jpg",
        "https://example.com/assets/lighting-ref.jpg"
      ],
      "resolution": "2k",
      "aspect_ratio": "4:3"
    }
  }
}

Referring to the images by their order in the prompt (image 1, image 2) makes the role of each one explicit. If an edit only needs one source image, sending one keeps the request at the base price.

Sizes and aspect ratios to ask for

Seedream 5.0 Pro output is shaped by three request fields, and the table lists the values this model accepts for each.

The same aspect ratios apply when editing. Prices differ by resolution, so state 1k or 2k when you ask and the quote will match. Pixel requirements for reference images come back in the image_size field of list_models for this model; the Seedream 5.0 Pro model page covers the rest of the specification.

Why was my Seedream 5.0 Pro image blocked by content moderation, and what gets flagged?

Makernice moderates every Seedream 5.0 Pro request at three points: the prompt is checked against a blocklist and automated moderation before it reaches the model, reference images are scanned before generation, and the finished image is scanned before it is shown or delivered. The Acceptable Use Policy lists what gets flagged: nudity and sexual content, graphic violence and gore, hate speech, anything that sexualizes or endangers minors, deepfakes or impersonation of real people, and protected characters, logos or brands used without permission. A prompt or reference image blocked before generation is not charged. A finished image withheld by output moderation is still charged, and support can review it by hand.

Why did my Seedream 5.0 Pro generation fail, and is a failed task charged?

A Seedream 5.0 Pro request is either refused before a job exists or fails after the job starts, and neither case is charged; the exception is a finished image blocked by output moderation. When Makernice declines a call, the agent receives a sentence of text and no job is created. The names in the first column are the refusal identifiers in Makernice's server code; the second column quotes the start of the returned text.

If an image job fails after starting, for example when moderation rejects a reference image, check_job reports status failed with an error object and billing released, and the held amount goes back to your balance. A finished image blocked by output moderation is still charged; failed generations has the details.