Back to browse
Burnish, a UI for any MCP server (no LLM, no chat)

Burnish, a UI for any MCP server (no LLM, no chat)

by BundyBear·Apr 22, 2026·3 points·2 comments

AI Analysis

●●SolidBig BrainNiche GemZero to One

Schema-to-UI inference for MCP servers when most lack any visual interface.

Strengths
  • Works with any MCP server without requiring custom UI builds
  • Infers visual components from JSON response shapes automatically
  • No LLM overhead for pure tool exploration and debugging
Weaknesses
  • MCP ecosystem still nascent with limited real-world adoption
  • Generic UI may not match purpose-built custom interfaces
Target Audience

Developers building with MCP servers, AI tool integrators

Similar To

MCP Inspector · Cursor MCP tools

Post Description

Burnish is a UI for MCP servers. You point it at a stdio or SSE endpoint and it renders each tool as an interactive form from the tool's JSON schema, then maps the response to a card, table, chart, or pipeline view based on the shape of the JSON. No LLM involved. The point is to see what a server actually does without wiring up a chat client or writing an inspection script.

MCP Apps (the official ext-apps extension) lets servers ship their own custom UI, but only if the server author builds one. Most don't. Burnish works with any server, including ones that have never heard of it, because it reads the schema and infers components from the output shape.

The hosted demo above runs a showcase server with 34 tools. Locally: `npx burnish -- npx -y @burnishdev/example-server`. Source is AGPL-3.0: https://github.com/danfking/burnish.

The thing I genuinely want to know is whether a generic explorer like this solves a real problem, or whether chat clients and MCP Apps already cover it well enough. Feedback welcome.

Similar Projects

AI/ML●●●Banger

AI Jam Sessions – MCP server that teaches AI to practice piano

Teaches LLMs to play piano by giving them ears, eyes, and persistent practice journals.

Zero to OneWizardryBig Brain
mikeyfrilot
103mo ago
Developer Tools●●●Banger

Overture – A visual plan interceptor for AI coding agents

Visual flowchart plans before code execution beats Cursor's text-only approach.

Solve My ProblemBig BrainSlick
PreciousH
313mo ago
AI/ML●●Solid

Open-source MCP server for AI podcast clipping

Audio energy + transcript heuristics suggest clips; hardware acceleration handles rendering.

Ship ItSolve My Problem
nsiradze
103mo ago
Developer Tools●●●Banger

MCP server that gives AI agents real interactive terminal sessions

Fixes the core gap: AI agents can now actually use interactive CLIs, not just dump output.

Solve My ProblemSlickShip It
sleepycoder21
203mo ago