Reprise
@tepesama
About Reprise
Reprise un-flattens a finished, flat AI design (ChatGPT/Midjourney/Claude) back into editable layers and rebuilds it bit-perfect — it returns a fidelity score (MAE 0.00 / 0 stray px) so you can verify, not just trust. Reproducing runs locally and free; remix one design into N on-
Basic information
Config
Add this server to your MCP-compatible client using the configuration below.
{
"mcpServers": {
"reprise": {
"type": "http",
"url": "https://tepesama-reprise-mcp.hf.space/gradio_api/mcp/http"
}
}
}Tools
No tools detected
We auto-extract tools from the README. The maintainer can list them under a ## Tools heading to populate this section.
Overview
What is Reprise?
Reprise converts a finished, flat AI design (from ChatGPT, Midjourney, Claude) back into editable layers and rebuilds it bit‑perfect, returning a fidelity score (MAE 0.00 / 0 stray px) so you can verify the result. It runs locally and free, and can remix a single design into multiple on‑brand variants.
How to use Reprise?
Open the web UI (https://tepesama-reprise-mcp.hf.space → “Try it” tab) and drop an image – no signup required. To add it as an MCP server, run:
claude mcp add --transport http reprise https://tepesama-reprise-mcp.hf.space/gradio_api/mcp/http
The available tools are: reproduce(image) → editable layers + fidelity score, autodetect(image) → detected elements, diagnose(original, reproduction) → fidelity metrics.
Key features of Reprise
- Turns flat AI designs into editable layers with fidelity scoring.
- Fidelity measured by Mean Absolute Error (MAE) and stray pixel count.
- Runs locally and is entirely free – no signup required.
- Remix one design into multiple on‑brand variants.
- Unedited layers remain bit‑perfect; only edited regions are newly generated.
- Auto element‑detection (v1) provides identified elements.
Use cases of Reprise
- Verify that AI‑generated designs are faithfully reproduced.
- Edit and remix a single AI design into several brand‑consistent variants.
- Integrate design verification and remixing into MCP‑based automation workflows.
FAQ from Reprise
How is fidelity measured?
Reprise returns a fidelity score with Mean Absolute Error (MAE) and stray pixel count (e.g., MAE 0.00 / 0 stray px) so you can quantitatively verify the reproduction.
What happens when layers are edited after reproduction?
Unedited layers remain bit‑perfect; only the regions you edit are newly generated, ensuring honesty in the output.
Is Reprise free to use?
Yes. Reproducing runs locally and is free, and you can try it immediately on the web UI without signing up.
What are the current limitations?
Auto element‑detection is version 1 and may not be perfect. The tool is designed to be honest by flagging what it cannot precisely reproduce.
What transport does the MCP server use?
The server uses HTTP transport, as shown in the provided claude mcp add command.
More AI & Agents MCP servers
21st.dev Magic AI Agent
21st-devIt's like v0 but in your Cursor/WindSurf/Cline. 21st dev Magic MCP server for working with your frontend like Magic
Open Multi-Agent Canvas
CopilotKitThe open-source multi-agent chat interface that lets you manage multiple agents in one dynamic conversation and add MCP servers for deep research
Sequential Thinking Multi-Agent System (MAS)
FradSerAn advanced sequential thinking process using a Multi-Agent System (MAS) built with the Agno framework and served via MCP.
meGPT - upload an author's content into an LLM
adriancoCode to process many kinds of content by an author into an MCP server
Mcp Agent
lastmile-aiBuild effective agents using Model Context Protocol and simple workflow patterns
Comments