Mockups Gallery

What Mockups Gallery MCP server does

The Mockups Gallery MCP server provides a persistent home for AI-generated design files. It accepts HTML and SVG as raw text, and PNG or JPG files as base64 content through the send_mockup tool. Uploaded files appear in a browser-based gallery and remain available on the server after the client removes its local copy.

The gallery organizes records by project, title, description, tags, content type, and timestamps. Developers can use it when generated screens, landing pages, or other visual drafts need a shared location outside a source repository. The web interface includes a project sidebar, a chronological feed, and a viewer for HTML and image content.

How it works

A single Docker container runs the FastAPI application. The MCP service is available through HTTP at /mcp/ and through Server-Sent Events at /mcp/sse; the gallery is served at /. MCP clients perform lightweight coordination tasks, while the application stores metadata in SQLite with WAL mode and files in a bind-mounted data/ directory.

For large mockups, the project provides POST /api/upload. An AI client can write a file locally and send it with curl, keeping the file body out of the model context. The MCP tools then manage the resulting record, metadata, and versions. Raw content can be read later by requesting the view_url returned by get_mockup.

Versions can be added explicitly with a parent design or inferred through folding. When an upload matches one existing design in the same project after supported title markers are removed, it may become another version automatically. Variant labels such as “option B” remain separate designs. Folded records retain aliases and old links where supported, and split_version can reverse an individual fold.

Setup and configuration

The Mockups Gallery MCP server requires Docker for deployment or Python 3.12 or newer for local development. It also requires an MCP-compatible client. A reverse proxy such as Traefik is optional and can provide TLS or authentication at the edge.

Persistent storage is the data/ directory beside the compose file. It contains the SQLite database and project-organized mockup files. The deployment has no built-in authentication: every reachable API endpoint and MCP tool is open to users who can access the server. Run it on a trusted LAN, VPN, or Tailscale network, or place it behind a proxy that supplies authentication before exposing it more broadly.

Tools and capabilities

Available MCP operations include:

• send_mockup uploads supported mockup content and returns a gallery URL.
• list_mockups lists records in reverse chronological order, with an optional project filter.
• get_mockup returns metadata, URLs, and the available version list for a UUID.
• update_mockup changes metadata or creates a new version when content is supplied.
• split_version extracts a version into a standalone mockup.
• delete_mockup removes a design or a selected version, including stored files.
• tag_mockup adds or removes tags.
• set_created_at changes the date for a design or version.

A separate one-time command can merge older duplicate designs after a dry run. It groups records by project and normalized base title, but bulk merging is not automatically performed.

Limitations and notes

The Mockups Gallery MCP server does not provide built-in authentication, so network placement matters. Its MCP tools are intended for metadata and coordination rather than transporting large file bodies; use the upload endpoint for that case. Deleting a version removes its file and may remove aliases pointing to it, while splitting a version changes the old version URL in some cases. Bulk folding is manual and should be reviewed before applying.