Releases·Tool Release
Prefect's Jeremiah Lowin ships Prefab, a Python DSL for generative UI on MCP apps
In a conference talk, FastMCP author and Prefect CEO Jeremiah Lowin introduces Prefab, a Python DSL for building generative UIs on top of MCP apps. MCP apps let tool results bypass the agent and render directly to users as HTML, CSS, and JavaScript, solving what Lowin calls the world's most expensive copy paste operation. Prefab targets enterprise Python engineers who need tables, forms, and charts rather than fully branded front ends.
98 engagement2 sourcesThu, Sep 10, 2026, 04:30 PM
AI Overview
- MCP apps, a 2026 protocol extension, let tool results bypass the agent and reach users as clickable HTML, CSS, and JavaScript interfaces
- Prefab is a Python DSL where nested context managers build UI structure and every component renders as a polished class, with reactive variables replacing JavaScript
- Lowin says the JSON representation in the Python-to-React pipeline is the real point: a serializable UI is one an agent can generate, receive, or edit
- Prefab's documentation renders entirely in Prefab and is editable in place
- UIs streamed as Python run roughly 70% smaller than the same UI as JSON, so Prefab now sends Python and converts it in a sandbox
- Without MCP apps, a file upload tool forces the agent to retype file contents character by character, which Lowin calls the world's most expensive copy paste operation
Related
Products·building
LLM generates MUI components in chat responses
A chatbot integration that allows an LLM to output Material-UI components (tables, charts, forms) directly in chat answers, eliminating the need to manually register each component. It solves the problem of static text-only responses by enabling dynamic, rich UI generation from the model.
A chatbot integration that allows an LLM to output Material-UI components (tables, charts, forms) directly in chat answers, eliminating the need to manually register each component. It solves the problem of static text-only responses by enabling dynamic, rich UI generation from the model.
8 engagement1 sourceredditTue, Jul 21, 2026, 08:39 AM
Products·launched
AI Prototyper: Figma plugin for GUI prototyping via LLM decomposition and RAG
AI Prototyper is an open-source Figma plugin that automates GUI prototyping from natural-language descriptions. It uses a decomposition and retrieval-augmented generation (RAG) pipeline with LLMs to break down a screen description into components and generate corresponding Figma elements. This reduces manual effort for designers and developers during early prototyping.
AI Prototyper is an open-source Figma plugin that automates GUI prototyping from natural-language descriptions. It uses a decomposition and retrieval-augmented generation (RAG) pipeline with LLMs to break down a screen description into components and generate corresponding Figma elements. This reduces manual effort for designers and developers during early prototyping.
0 engagement1 sourcearxivThu, Jul 16, 2026, 10:49 AM
Releases·Tool Release
MCP Apps: MCP servers return interactive UIs instead of text
Liad Yosef, co-creator of MCP UI, describes MCP Apps, a mechanism for MCP servers to return interactive interfaces instead of text blocks. Built on MCP UI and shaped through an open working group in the MCP committee, it lets hosts render web components and route clicks back into the agentic loop.
Liad Yosef, co-creator of MCP UI, describes MCP Apps, a mechanism for MCP servers to return interactive interfaces instead of text blocks. Built on MCP UI and shaped through an open working group in the MCP committee, it lets hosts render web components and route clicks back into the agentic loop.
58 engagement1 sourceyoutubeSun, Aug 2, 2026, 11:30 PM
Releases·Tool Release
Figma engineer recounts launching Figma's first MCP server as a one-day-a-week side project
In a talk, Figma's Jesse Lumarie describes how the company's first MCP server grew from an unofficial one-day-a-week project into one of Figma's fastest-growing products in about three months. The build navigated a shifting MCP spec and uneven client support. It matters to practitioners as a case study in shipping MCP integrations against an unstable protocol.
In a talk, Figma's Jesse Lumarie describes how the company's first MCP server grew from an unofficial one-day-a-week project into one of Figma's fastest-growing products in about three months. The build navigated a shifting MCP spec and uneven client support. It matters to practitioners as a case study in shipping MCP integrations against an unstable protocol.
1 engagement1 sourceyoutubeThu, Aug 27, 2026, 12:02 PM
Products·building
Vendo - open-source layer for AI agents to render real UI components, not markdown
Vendo is an open-source layer that lets LLM agents render real UI (charts, forms, settings panels) inside a product by picking from a fixed set of pre-declared components. It solves the problem of agents returning only markdown text and requiring custom glue per UI case. Aimed at developers building agent-driven products.
Vendo is an open-source layer that lets LLM agents render real UI (charts, forms, settings panels) inside a product by picking from a fixed set of pre-declared components. It solves the problem of agents returning only markdown text and requiring custom glue per UI case. Aimed at developers building agent-driven products.
2 engagement1 sourceredditThu, Sep 3, 2026, 09:17 PM