MCP

MCP

The DesignerBox MCP server lets an AI assistant work directly with your DesignerBox account. Ask Claude to generate a product shot, apply a template, check your credit balance, or pull an asset out of your library, and it happens in your account and shows up in the app.

MCP (Model Context Protocol) is an open standard for connecting AI assistants to external tools. DesignerBox runs a hosted MCP server, so there is nothing to install and nothing to self-host.

Server URL

Connect your client to:

https://mcp.designerbox.ai/api

The path is /api, not /mcp. The server speaks JSON-RPC 2.0 over streamable HTTP.

Connect your assistant

The exact steps differ per client, but the shape is always the same: add a custom connector, paste the server URL, and sign in.

Claude (web, Desktop, and Cowork)

1. Open Settings, then Connectors.
2. Choose to add a custom connector.
3. Paste https://mcp.designerbox.ai/api.
4. Complete the DesignerBox sign-in when your browser opens.
5. Start a chat and ask for something, for example "list my DesignerBox designs".

The connector registers itself automatically, so there is no client ID or secret to copy. Once connected in Claude, the same connection is available in Cowork.

Cursor and other MCP clients

Any client that supports remote MCP servers over streamable HTTP works. Point it at the same URL. Clients that support OAuth discovery will find the sign-in flow on their own.

Authentication

DesignerBox MCP uses OAuth only. You sign in through your browser, and your client receives a token scoped to your account. There is no API key to paste, and API keys from other parts of DesignerBox are not accepted here.

The server supports dynamic client registration, so well-behaved clients configure themselves from the URL alone. Tokens are issued with these scopes:

• openid - identifies your account.
• mcp:read - read your designs, assets, templates, and plan.
• mcp:write - generate media and change your data.
• offline_access - refresh the session so you do not sign in every day.

Your team membership is checked on every request, so removing someone from a team revokes their access immediately, without waiting for a token to expire.

To review or revoke a connected assistant, open MCP connections in your account settings in the app. Revoking there disconnects that client.

What your assistant can do

The server exposes 63 tools, 37 that read and 26 that write. You do not need to learn their names. Ask for what you want in plain language and the assistant picks the tool.

The main areas:

• Generate. Images, video, cinema studio shots, and AI characters.
• Edit. Background removal, relighting and angles, skin retouch, virtual try-on, flat lays, and product ads.
• Templates and techniques. Browse the curated catalog and apply a template to your own reference images.
• Library. List and fetch your designs, assets, folders, and galleries, or upload a new asset straight from the chat.
• Brand. Read and update your brand profile and brand assets.
• Workflows. List, create, and run pipelines from the canvas.
• Account. Check your plan, credit balance, limits, and teams.

Read tools and write tools are labeled so your assistant knows which actions change your account. Most clients ask you to confirm before running a write tool the first time.

Try these

• "Generate a product photo of a matte black water bottle on a concrete surface."
• "What templates do you have for fashion flat lays?"
• "How many credits do I have left this month?"
• "Remove the background from the last image you made."
• "List my designs from this week."

Plans and credits

Generations spend credits from the same monthly pool the app uses. There is no separate MCP allowance and no extra charge for using MCP. The Free plan includes 112 credits (about four images on the default model), so you can try generation before upgrading; once credits run out, generation tools stop until your allowance refreshes or you upgrade, while the read tools keep working.

Image cost depends on the model:

Other operations:

Ask your assistant to list the image or video models if you want current costs and capabilities without leaving the chat. See Plans, Credits, and Limits for the full picture.

How long things take

Some tools finish in the conversation, others run in the background:

• Images finish in the chat. A generation takes roughly 30 seconds, and the assistant returns the image when it is done.
• Video and AI characters run in the background. You get a task reference immediately, and the assistant checks the status until it is ready. You can keep working, or come back later and ask whether it finished.

Anything you generate is saved to your DesignerBox library, so it is in the app whether or not the chat is still open.

Limits

Requests are rate limited per account. Normal conversational use stays well inside the limits. If an assistant loops or fires a large batch, it may be throttled briefly and should retry.

Troubleshooting

The assistant says it cannot find the server. Check the URL ends in /api. https://mcp.designerbox.ai/mcp is not valid and returns not found.

Sign-in succeeds but tools fail with an authorization error. Your token may have been revoked, or you were removed from the team. Reconnect the connector.

A generation stops and mentions credits. Your monthly credits are spent (the Free plan starts with 112). Read tools keep working; to generate again, wait for your allowance to refresh, buy a credit pack, or upgrade. Check your balance in the app, or ask the assistant for your plan.

The assistant picks the wrong tool. Be specific about the outcome. "Apply the linen flat lay template to this photo" beats "make this nicer".

Related

• API - the REST API, for code rather than conversation.
• Plans, Credits, and Limits - how generations consume credits.
• Workflows and the Canvas - the pipelines MCP can run.
• Techniques and Templates - the catalog your assistant can apply.