GuidesHow to Use Seedream 5.5 in Claude Code and Cursor?
Makernice Team · ·
Prices as of November 12, 2026 (price list v15).
Key takeaways
- Seedream 5.5, ByteDance's image model, is available to Claude Code and Cursor through the Makernice MCP server.
- Seedream 5.5 takes a background field with auto, transparent or opaque, so an agent can ask for a cut-out asset in one step.
- Seedream 5.5 edits accept up to 10 reference images, and every reference image, the first included, adds a fixed amount to the edit price.
- Output comes at 1k or 2k and is billed per image; estimate_cost returns the exact total before you approve.
Seedream 5.5 can be used from Claude Code or Cursor once the Makernice MCP server is added: the agent asks estimate_cost for the price, you approve it, and generate makes the image. At price list v15 (November 12, 2026) a 1k text-to-image result is $0.0156.
Adding Makernice to your editor's agent
Cursor: install from the Cursor connect page with one click, or paste its snippet into ~/.cursor/mcp.json. Sign in next to makernice in Cursor's MCP settings and approve the browser prompt.
Claude Code: the Claude Code connect page shows the claude mcp add --transport http --scope user makernice command with the server URL filled in. After running it, start Claude Code, type /mcp, choose makernice and log in. Ask for get_balance afterwards to confirm the tools are loaded and to see the connection's per-job limit.
Does the Seedream 5.5 flow change in Claude or ChatGPT?
No, only the setup does. A Seedream 5.5 job goes through identical tool calls in every client: an estimate_cost quote that informs you but locks nothing, a generate call capped by max_cost_usd, then check_job until the image is ready.
- In Claude, open Customize, then Connectors, and choose Add custom connector; paste the MCP URL and connect with your Makernice login. Team and Enterprise workspaces need an owner to add it first. Full steps: Claude connect page.
- In ChatGPT on the web, open Plugins and add a custom MCP server with OAuth, pasting the ChatGPT-specific address; your plan or workspace must allow this. Mention Makernice with @ in a new chat to use it. Full steps: ChatGPT connect page.
Transcript: a transparent-background asset from Seedream 5.5
You: Make a square sticker of a fox astronaut with a transparent background for the app's empty state. Tell me the price.
Claude Code adds background to the request and calls:
{
"name": "estimate_cost",
"arguments": {
"input": {
"model": "seedream-5.5",
"mode": "t2i",
"prompt": "Sticker of a cartoon fox astronaut holding a tiny flag, thick white outline, flat colors, centered",
"resolution": "1k",
"aspect_ratio": "1:1",
"background": "transparent"
}
}
}Claude Code: Seedream 5.5 quotes $0.0156 for one 1k image with a transparent background (price list v15). The estimate didn't create anything. Make it?
You: Yes.
{
"name": "generate",
"arguments": {
"input": {
"model": "seedream-5.5",
"mode": "t2i",
"prompt": "Sticker of a cartoon fox astronaut holding a tiny flag, thick white outline, flat colors, centered",
"resolution": "1k",
"aspect_ratio": "1:1",
"background": "transparent"
},
"max_cost_usd": "0.0156",
"request_id": "fox-astronaut-sticker-alpha"
}
}Makernice prices the input again the moment generate arrives and only creates the job if the new figure is no higher than max_cost_usd and fits the per-job limit; the charge equals that figure. A reply with job_id follows. A retry carrying the same request_id and identical input brings back that job rather than a second one.
Then check_job with a wait_s of up to 55 seconds waits for the result. On success, outputs[0].url is a signed download link good for 24 hours, content_type gives the file type, and a later check_job call refreshes the link. The run is also listed at console_url, and an account webhook fires for it if you have one set.
The background option and the other Seedream 5.5 fields
Seedream 5.5 requests use four optional fields, and the table lists what each one accepts on this model.
| Field | Seedream 5.5 accepts | Notes |
|---|---|---|
resolution | 1k or 2k | 1k when left out; each has its own per-image rate |
aspect_ratio | 1:1, 4:3, 3:4, 16:9, 9:16, 2:3, 3:2, 21:9 | Same list for text-to-image and edits |
background | auto, transparent, opaque | auto is the default; transparent suits stickers, icons and product cut-outs |
reference_images | 1 to 10 https links | Edits only (mode i2i) |
The 2k rate is $0.0273 per text-to-image result. Reference images need to be public https links to JPEG, PNG or WebP files of 10 MB or less, and the pixel rules come back in image_size when the agent calls list_models with this model's id. The Seedream 5.5 model page has the full specification.
How do I edit images with Seedream 5.5, and can it do layers or inpainting?
Seedream 5.5 edits run in image-to-image mode: set mode to i2i, pass 1 to 10 public https links in reference_images, and write the change in prompt, for example a new outfit for the character in the first image. Seedream 5.5 has no mask or layer field, so inpainting a selected area and layer-by-layer output are not supported; describe the area you want changed in the prompt. The background field works on edits too, so an edit can return the subject on a transparent background.
Paying per reference image when Seedream 5.5 edits
An edit costs a base amount plus a charge for each reference image. The base is $0.0156 at 1k and $0.0273 at 2k, and each reference adds $0.00624, starting with the first. So the number of links in reference_images changes the price, and the agent should quote again whenever it adds or drops one. list_models shows the add-on as per_reference_price_usd.
Here the agent quotes a 2k edit with two references, keeping the transparent background:
{
"name": "estimate_cost",
"arguments": {
"input": {
"model": "seedream-5.5",
"mode": "i2i",
"prompt": "Redraw the fox from the first image wearing the scarf pattern from the second image, same sticker style",
"reference_images": [
"https://example.com/stickers/fox-astronaut.png",
"https://example.com/patterns/plaid-scarf.jpg"
],
"resolution": "2k",
"background": "transparent"
}
}
}Seedream 5.5 errors and the fix for each
A refusal arrives as a line of text; no job is created and nothing is charged. Below, the first column is the refusal name used inside Makernice's server and the second is the start of what the tool sends back.
| Refusal | What the tool returns | What to do |
|---|---|---|
price_exceeds_max_cost | "which exceeds max_cost_usd" | An extra reference image or a switch to 2k pushed the price up; estimate again and approve the new total |
invalid_request | "Invalid parameters" | The message names the bad field, for example more than 10 reference images; compare it with list_models |
per_job_limit_exceeded | "which exceeds this connection's per-job limit" | Use fewer references or 1k, or raise the per-job limit for this connection |
key_spend_limit_reached | "This connection has reached its spend limit. Nothing was charged." | Raise the connection's total spend limit in Makernice |
Jobs that fail after they start, for instance on a reference image that moderation rejects, end with status failed and an error object in check_job; billing changes to released and the held amount is returned. A finished image that output moderation blocks is still charged, as explained on failed generations.
