Self-hosted Penpot MCP alternatives
Penpot's MCP server can run locally over stdio, and Penpot itself is built to self-host, so the design files and the server process can both stay on infrastructure you control. The server reads a file's overview, inspects and exports shapes, imports images, and runs plugin code.
The servers below also install locally. Framelink reads design files from a local process; the rest are general-purpose tools you can run yourself to round out a design workflow, from pulling references to holding context to managing other servers. None of them is a hosted-only product.
The 8 best self-hosted alternatives
Framelink's Figma server runs locally and feeds clean layout and styling context from a Figma file to a coding agent, with tools to read Figma data and download images. The local choice when your designs are in Figma.
Set up Framelink Figma MCP →Adjacent to design: the reference Fetch server runs locally and converts a URL to clean Markdown. It pulls reference material or specs into an agent's context from a process you control.
Set up Fetch →The reference Memory server runs locally and gives an agent a persistent knowledge graph of entities, relations, and observations, holding design decisions across a session on your own machine.
Set up Memory →Small and self-contained, the reference Time server runs locally for current-time lookups and timezone conversion from the IANA database, supporting scheduling around a design workflow.
Set up Time →Run it yourself and the community Slack server reads history, DMs, and search over stdio with no workspace bot. It reaches design feedback that happens in chat rather than in a file.
Set up Slack →For tracking design tasks or assets, the Airtable server installs locally and is schema-first: inspect a base, then read, search, and write records, fields, and comments from a process you operate.
Set up Airtable →The Coda server runs on your own machine and connects an agent to docs, pages, tables, and rows. It fits design briefs and spec docs kept in Coda alongside the canvas work.
Set up Coda →- DockerOfficial
Docker's MCP Gateway runs locally to run, secure, and aggregate other containerized MCP servers behind one endpoint. Useful for operating a self-hosted design toolchain, not for editing designs itself.
Set up Docker →
How to choose
Penpot self-hosts the canvas and its files; this group adds local tools around it. Framelink reads Figma files from a process you run, while Fetch, Memory, Time, Slack, Airtable, and Coda cover references, context, chat, and project docs. Docker's Gateway manages the rest. One caveat: running a server locally keeps its process and credentials on your machine, but services like Slack, Airtable, and Coda still send data to their own APIs.
FAQ
- Can the Penpot MCP server be self-hosted?
- Yes. Penpot's server runs locally over stdio, and Penpot itself is designed to self-host, so the design files and the server process can both stay on infrastructure you control. The server reads file overviews, inspects and exports shapes, imports images, and runs plugin code.
- Are these self-hosted design tools?
- Mostly they are general-purpose. Framelink Figma is the one that reads design files, from a local process. Fetch, Memory, Time, Slack, Airtable, Coda, and Docker run locally to fetch references, hold context, read chat, track tasks, or manage other servers, so they support a design workflow rather than edit the canvas.