AI assistants (MCP)
Connect On-Model to Claude, ChatGPT, Cursor and other AI assistants with the On-Model MCP server
Connect On-Model to Claude and other AI assistants. With the On-Model MCP server you ask in plain language and the assistant runs On-Model for you: it can put flat-lay garments on a model, make packshots, recolor products, swap models and run whole collections.
| Server URL | https://mcp.on-model.com/mcp |
| Transport | Streamable HTTP |
| Authentication | OAuth: you sign in with your On-Model account, no API key needed. API tokens work too (see Claude Code) |
Requirements
- An On-Model account on a Pro or Enterprise plan. Connecting an assistant uses API access, which those plans include. Free and Basic accounts can't connect (pricing).
- Credits for generation. Browsing models, templates and results is free.
Signed up with Google?
Accounts created with Google sign-in have no password yet. Set one with Forgot password (Google sign-in keeps working), or connect with an API token from your profile under API tokens.
Connect
Claude (claude.ai, Desktop, mobile)
- Open Settings → Connectors.
- Click Add custom connector and paste
https://mcp.on-model.com/mcp. - Click Connect, sign in with your On-Model email and password, and you're sent back to Claude.
- In a chat, make sure On-Model is enabled in the tools menu.
ChatGPT
Add a connector with the server URL https://mcp.on-model.com/mcp and sign in when asked. Custom connectors are available on some ChatGPT plans and may need to be enabled in ChatGPT's settings first.
Claude Code
claude mcp add --transport http on-model https://mcp.on-model.com/mcpThen run /mcp and sign in when prompted. For scripts or CI, pass an API token instead of signing in:
claude mcp add --transport http on-model https://mcp.on-model.com/mcp \
--header "Authorization: Bearer YOUR_API_TOKEN"Other MCP clients (Cursor and others)
Add a remote server with the URL above and the Streamable HTTP transport. The client discovers the sign-in endpoints automatically. Clients without OAuth support can send an API token as Authorization: Bearer <token>.
Example prompts
| Ask | What happens |
|---|---|
| "Here are flat-lays of a linen shirt, chinos and loafers. Put the outfit on a male model in his 30s, full body, light grey studio, 2K." | Uploads the photos, picks a model, generates the on-model image |
| "Make a ghost-mannequin and a white-cutout packshot of this jacket." | Creates two packshots from your product photos |
| "Recolor this dress in forest green, keep everything else identical." | Creates a new colourway of the same images |
| "Replace the model in these product photos with a dark-skinned model, and keep the clothes exactly as they are." | Model swap, keeping the photo's own clothes and background |
| "Run my 'Summer PDP' template on all 40 tagged products, and show me the cost first." | Runs a free dry run with the credit estimate, then generates once you confirm |
| "Build a template in the style of these four photos from our webshop." | Saves a reusable template matching your house style |
Tools
| Tool | What it does | Credits |
|---|---|---|
upload_images | Upload garment or product photos (URL or base64) | Free |
flat_to_model | Dress a model in flat-lay or packshot garments | 3 / 5 / 10 per image (1K / 2K / 4K) |
create_packshot | Ghost mannequin, flat lay, white cutout or marketing scene | 3 / 5 / 10 per image |
garment_recolor | New colourways of one garment | 3 / 5 / 10 per image (low / medium / high quality) |
model_swap | Re-cast existing photos with another model; can keep the photo's clothes and background identical | 5 / 7 / 10 per image (low / medium / high quality) |
image_playground | Free-prompt generation and edits | 3 / 5 / 10 per variation |
bulk_create | Run one setup across up to 100 products (dry run first) | As per job type |
fix_images | AI retouch of images the reviewer flagged (Pro) | 1 per fixed image |
create_identity, promote_identity | Generate new synthetic models; keep the ones you like | Per draft; 50 to keep |
add_identity | Digital twin of a consenting brand model from one photo | 50 |
create_template_from_images | Template in a brand's style from 2–6 photos | 1 per image |
create_template, duplicate_template, update_template, create_preset, update_preset, add_preset_to_template | Manage templates and presets | Free (creating a preset from a description or image uses credits) |
list_identities, list_templates, list_presets, list_tags, get_template, get_job, get_download_link, get_credits, identity_status, share_identity | Look up models, templates, jobs, your plan and credits | Free |
Good to know
- Costs are confirmed first. The assistant is instructed to state the credit cost and ask before large jobs;
bulk_createalways runs a free dry run first. - Jobs take about a minute. The assistant checks the result and shows previews in the chat. Full-size images and ZIP downloads come as links.
- Color accuracy. Finishing looks change garment colors too. Ask for no look when exact color matters.
- People in photos. Only create digital twins of people who have agreed to it.
- Jobs appear in your account like any other job, so you can review and download them in the app too.
Privacy and security
On-Model receives only what the assistant sends when it calls a tool (images, settings, instructions), never your conversation. Your password is used once to create an access token for the connection and isn't stored. You can see that token in your On-Model profile under API tokens. To disconnect, remove the connector in your assistant, or revoke the token in your profile, which disconnects every assistant using it. Full details: privacy policy.
Support
Questions or problems: office@piktid.com.