What is the mockup-gallery MCP server?
mockup-gallery, built as mockups-mpc, is a self-hosted MCP server and gallery for AI-generated mockups. An agent writes a mockup locally, then uploads it with a lightweight curl call instead of passing the whole thing through the model's context, while a FastAPI and SQLite backend stores it for a web gallery to browse.
This server gives AI-generated mockups, such as HTML, SVG or image exports, a lasting, self-hosted home instead of leaving them as text embedded in a chat or lost once a session ends.
An agent writes a mockup to a local path, then uploads it with curl to the server's endpoint. The MCP tools themselves stay lightweight, handling listing, tagging and metadata rather than large uploads.
What it changes in your workflow
- Mockups get a permanent address and a web gallery to browse, instead of disappearing when a chat session ends.
- Uploads go through a token-efficient curl call, keeping large mockup content out of the model's own context.
- A mockup can be split into a standalone version once a variant is worth keeping on its own.
- Tags and project grouping make a growing mockup collection searchable later.
How it stores mockups
| mockup-gallery | |
|---|---|
| Backend | FastAPI plus SQLite |
| Upload path | curl against a local upload endpoint |
| Browsing | A web gallery served by the same project |
| Setup | docker compose up, pointed at a local compose definition |
Who it is for
Best for
- Teams generating many AI mockups who need somewhere lasting to keep and compare them.
- Agents that want mockup storage handled without spending context tokens on upload content.
- Projects that want tagging and versioning on top of plain mockups rather than a flat list.
Not for
- Writing the HTML mockup content itself, rather than storing one already produced, is a separate step that the Baoyu Design skill handles for prototypes and decks.
- Building and editing vector artwork in a self-hosted Penpot project, rather than storing a finished mockup export, is the job the connected MCP server does for an agent.
- Uploading a design system document and applying its tokens across a whole project, rather than storing a finished mockup, is the job Stitch Manage Design System does.
- Looping through a whole site build one page at a time from a brief, instead of storing mockups already produced, is what the Design Loop skill does.
Use it with UiChemy
This server keeps a record of mockups an agent already produced. Once one is picked, UiChemy Composer or AI to WordPress can turn it into an editable Elementor or Gutenberg page. Click Get Started on uichemy.com.
Connect Mockup Gallery MCP Server: Store and Browse AI Mockups to your AI client
docker compose -f docker-compose.local.yml up -dClone and run the compose stack
Run docker compose -f docker-compose.local.yml up -d from the project to start the FastAPI backend and MCP server together.
Add the server entry
Use the SSE config above in your desktop client, pointed at your own server's address.
Upload a mockup
Have the agent write a mockup locally, then curl it to the server's upload endpoint instead of sending it through chat.
Browse the gallery
Open the project's own web gallery in a browser to review, tag or compare stored mockups.
Client configuration
{
"mcpServers": {
"mockups-gallery": {
"type": "sse",
"url": "https://your-domain.example.com/mcp/sse"
}
}
}What you can do with it
- Token-efficient uploadsA curl upload keeps the mockup's actual content out of the model's context window, with the MCP tools handling only lightweight metadata calls.
- Permanent storageFastAPI and SQLite give every mockup a lasting address instead of one tied to a single chat session.
- Version splittingA promising variant can be split into its own standalone mockup rather than staying buried as one version among many.
- Browsable web galleryThe same project serves a gallery so a person can review and compare mockups without going through the agent.
- Tagging and project groupingMockups can be tagged and grouped by project, which keeps a growing collection searchable.
- Docker-first setupThe documented setup runs the backend and MCP server together through a provided compose definition.
Tools it exposes
send_mockup
Registers an uploaded HTML, SVG, PNG or JPG mockup with the gallery after it reaches the upload endpoint.
list_mockups
Lists stored mockups, with optional filtering by project.
get_mockup
Returns details for one mockup by its id, including its stored versions.
update_mockup
Updates a mockup's metadata or attaches a new version to it.
split_version
Separates one version of a mockup into its own standalone entry.
delete_mockup
Removes a mockup, or a single version of one, from the gallery.
tag_mockup
Adds or removes tags on a mockup for later search and grouping.
set_created_at
Adjusts a mockup's recorded creation time, useful when backfilling older work.
Frequently asked questions
What is the mockup-gallery MCP server?
It is a self-hosted MCP server and web gallery for AI-generated mockups, built as mockups-mpc. It stores mockups uploaded through a token-efficient curl call and exposes 8 tools for listing, tagging and managing them.
How do I install the mockup-gallery MCP server?
Clone the repository and run docker compose -f docker-compose.local.yml up -d to start the FastAPI backend together with the MCP server, then connect a desktop client over the server's SSE endpoint.
Why does it use curl instead of sending the mockup through the MCP tools?
Passing a whole HTML or image mockup through the model's context is expensive in tokens. The agent writes the mockup locally and uploads it with curl, so the MCP tools only handle lightweight metadata operations.
Can I browse stored mockups outside the agent?
Yes. The project serves its own web gallery, so a person can open it in a browser and review, tag or compare stored mockups without going through the agent at all.
What does split_version do?
It separates one version of a mockup into a standalone entry of its own, which matters when a variant turns out good enough to keep developing on its own path.
Which content types can it store?
The send_mockup tool accepts HTML, SVG, PNG and JPG content, covering both code-based mockups and rendered image exports side by side in the one shared web gallery view, with tags on each entry.
Does it need a cloud account?
No. The project is self-hosted with a FastAPI and SQLite backend, run through Docker on infrastructure you control, with no external account required to store, tag or browse mockups at any point.


