io.github.QuasarG/taste-bank MCP Server
io.github.QuasarG/taste-bank
Reusable visual design systems for AI coding agents — browse, fetch, and submit styles via CLI and MCP.
What is the io.github.QuasarG/taste-bank MCP server?
The Taste Bank MCP server is a design-style library and distribution platform for AI coding agents like Claude, Cursor, and 70+ others. It lets agents browse, fetch, and apply pre-built front-end style packs (design tokens, CSS, templates, and usage guides) to produce well-designed interfaces consistently. Developers can also submit and version their own styles via CLI with ed25519-signed ownership.
Taste Bank solves the problem of getting coding agents to produce well-designed front-ends by providing a unified, reusable library of structured style packs. Each pack includes design tokens, scoped CSS, HTML templates, and a skill guide. Agents can browse styles visually at tastebank.cloud, fetch complete packs via CLI or MCP, and apply them to projects. Maintainers can submit and version their own styles with cryptographic ownership, and the community votes with usage — only truly useful styles gain traction.
How to install io.github.QuasarG/taste-bank
Copy-paste configuration for popular MCP clients.
No machine-readable install method is published for this server in the registry. Check the repository or website for setup instructions.
Tools & capabilities
Tools this server exposes to the agent.
taste-bank list— Browse and search published styles with optional keyword filtering and JSON outputtaste-bank show— Retrieve style metadata and design tokens for a specific style slugtaste-bank skill— Fetch the complete style pack (meta, tokens, skill guide, scoped CSS, and HTML templates) in JSON or Markdown formattaste-bank use— Land a style into your project as a managed rules file without clobbering existing editstaste-bank favorite / unfavorite / favorites— Manage a local list of favorite stylestaste-bank doctor— Health check for CLI, network, identity, skill injection, and local cachetaste-bank keygen— Generate an ed25519 keypair for style ownership and submission signingtaste-bank whoami— Display bound identity, owned styles, and pending submissionstaste-bank validate— Dry-run validation of a style pack JSON against the zod schemataste-bank submit— Sign and submit a new style pack (requires invite code and private key)taste-bank update— Update an owned style with a new version (requires signature and version bump)taste-bank delete— Delete an owned style (requires confirmation and signature)
Use cases
- Browse and apply pre-built front-end design styles to AI-generated code for consistent, professional-looking interfaces
- Submit and version your own custom design styles with cryptographic ownership so they can be reused across projects and agents
- Teach coding agents (Claude, Cursor, etc.) to use design systems by injecting the taste-bank skill via one-line setup
- Fetch complete style packs including design tokens, CSS, templates, and usage guides for agents to implement strictly within constraints
- Manage and discover community-contributed styles via a TikTok-style gallery with leaderboards and live-rendered previews
io.github.QuasarG/taste-bank MCP server FAQ
Taste Bank is a design-style library and distribution platform for AI coding agents. It lets agents browse, fetch, and apply pre-built front-end style packs (design tokens, CSS, templates, usage guides) to produce well-designed interfaces. Developers can also submit and version their own styles with cryptographic ownership.
Run `npx taste-bank setup` once. This installs the CLI globally and injects two skills into your agents (Claude Code, Cursor, Codex, ZCode, Kimi Code, and 70+ others via the open skills ecosystem). After setup, you can tell your agent 'use a taste-bank style' and it will browse, fetch, and apply styles automatically.
Yes. Browsing and applying styles is free and requires no authentication. Submitting your own styles requires an invite code (contact the maintainer), but once you have one, submission and updates are free.
Run `taste-bank keygen` to generate an ed25519 keypair, add your invite code to `~/.style-lab/config.json`, then run `taste-bank submit <pack.json>`. Your submission enters a review queue and goes live after maintainer approval. You can also tell your agent 'submit this project's style to taste-bank' and the injected skill will guide it through the full process.
Browsing and applying styles requires no authentication. Submitting styles requires an invite code (binds to your public key on first use) and an ed25519 private key for signing submissions. Your key is your identity — no accounts or passwords.
Yes. Set the `TASTEBANK_API` environment variable to your own host URL. The CLI and MCP server will use that endpoint instead of the default tastebank.cloud.
README (reference)
Source of truth, from the repository.
Why Taste Bank
Sound familiar?
- Staring at a blank page, no idea how to lay out your components?
- Layout finally settled — yet the AI's visual style never quite lands?
- A hoard of "beautiful designs" you can never actually summon when you need one?
- The style you painstakingly tuned in your last project has to be re-explained from scratch in the next one?
Getting coding agents to produce well-designed front-ends is a solved problem nobody can actually use. The community is full of front-end skills, prompt snippets, and design guides — but in practice:
- No unified invocation: some styles ship as a skill, some as a plugin, some ship nothing at all. Actually using them is a chore.
- No quality bar: without a shared schema, a "style" can be three adjectives in a trench coat — nothing an agent can actually execute.
- Hard to reuse: great styles die in chat histories instead of being distilled into an asset.
- No ownership or iteration: no versions, no way to say "only I maintain my styles."
More importantly: the web spawns new styles every day — but not every style is mainstream enough, not every style is your taste. We want a community where people share the front-end taste they actually use — and vote with their feet. Only what truly gets used is truly usable.
Taste Bank's answer: distill each style into a structured style pack (SKILL.md usage guide + precise design tokens + templates/ snapshots), enforced by a zod schema, served through a CLI + skill injection that works with any coding agent. Once a style is in the bank, it's reachable anywhere, versioned, and owned.
Quick Start
Via CLI (recommended)
One command — installs the CLI globally and teaches your coding agents (Claude Code, Codex, Cursor, ZCode, Kimi Code…) how to use it:
npx taste-bank setup
Then just talk to your agent in plain language:
"Build me a landing page with one of the styles in taste-bank."
Your agent already knows what to do — setup injected a skill that teaches it the taste-bank CLI commands. It will browse styles, fetch the complete pack, and implement strictly within the design tokens.
Prefer the terminal?
taste-bank list # browse all styles
taste-bank skill <slug> # get the complete pack (meta + tokens + skill + css + templates)
taste-bank use <slug> # land a style into your project as a rules file
No per-agent config, no JSON to paste. Browse visually first at tastebank.cloud.
Via MCP (legacy)
For agents that prefer the MCP protocol, configure one URL in your MCP client:
{
"mcpServers": {
"taste-bank": {
"url": "https://tastebank.cloud/mcp",
"headers": { "x-invite-code": "sl_your_invite_code" }
}
}
}
Both paths call the same backend. CLI is recommended — simpler, works with 70+ agents out of the box.
Features
- One-line setup:
npx taste-bank setupinstalls the CLI and injects skills into all your coding agents (Claude Code, Codex, Cursor, ZCode, Kimi Code, and 70+ more via the openskillsecosystem) - Complete style packs:
taste-bank skill <slug>returns the full pack in one call — meta, tokens, SKILL.md rules, scoped CSS, and HTML template snapshots. Agents get everything they need. - 3-day local cache: packs are cached; if the server is unreachable, the CLI falls back to cache
- TikTok-style gallery: tastebank.cloud — endless stream of live-rendered previews, horizontal card rail, leaderboards
- Structured & validated: every pack passes a strict zod schema (meta / tokens / skill / templates), versioned, with invite-only submissions + ed25519-signed ownership + human review
- Submit from the terminal:
taste-bank submit— sign and submit a new style without leaving your project - Private key as identity: no accounts — whoever holds the key manages the style
CLI Commands
Browsing & applying (no auth)
| Command | What it does |
|---|---|
taste-bank list [--q WORD] [--json] | List published styles |
taste-bank show <slug> [--json] | Style detail (meta + design tokens) |
taste-bank skill <slug> [--md] | Complete pack as JSON: {meta, tokens, skill, css, templates}. --md for plain SKILL.md |
taste-bank use <slug> [--as agents|claude|skill] | Land a style into your project (managed sentinel block, won't clobber your edits) |
taste-bank favorite <slug> / unfavorite / favorites | Manage favorites |
taste-bank doctor | Health check (CLI, network, identity, skill injection, cache) |
Contributing (invite + ed25519 identity)
| Command | What it does |
|---|---|
taste-bank keygen | Generate an ed25519 keypair into ~/.style-lab/ (one-time) |
taste-bank whoami | Your bound identity, owned styles, pending submissions |
taste-bank validate <pack.json> | Dry-run validate a pack locally (no network) |
taste-bank submit <pack.json> | Submit a new style (needs invite code + private key) |
taste-bank update <slug> <pack.json> | Update an owned style (version must bump) |
taste-bank delete <slug> | Delete an owned style (asks to confirm) |
How it works
npx taste-bank setup
① npm install -g taste-bank (install CLI globally)
② npx skills add …/taste-bank -g (inject 2 skills into your agents:
taste-bank for browsing/applying,
taste-bank-contribute for submitting)
③ environment check (report state, read-only)
Two skills are injected (via the open skills ecosystem):
taste-bank— teaches your agent the CLI commands for browsing and applying stylestaste-bank-contribute— the full contribution SOP: how to sample a project, pack a style, sanitize it, and submit
After setup, your agent knows the CLI. You just say "use a taste-bank style" and it runs the commands itself.
The CLI talks to tastebank.cloud by default and caches packs locally for 3 days. Self-host? Set TASTEBANK_API=https://your-host.
Submit your own style
# One-time identity setup
taste-bank keygen
echo '{ "inviteCode": "sl_xxx" }' > ~/.style-lab/config.json # get a code from the maintainer
# From any project with a style worth keeping:
taste-bank validate my-style.pack.json # dry-run check
taste-bank submit my-style.pack.json # sign + submit (enters review queue)
Or just tell your agent: "submit this project's style to taste-bank" — the injected taste-bank-contribute skill guides it through the full SOP (sample 2 files max, extract real tokens, sanitize business-identifying content, pack, validate, submit).
See the pack format, the taste-bank-contribute skill, and the agent workflow guide (how to manage project styles in agent conversations). 中文版见 Agent 工作流指南。
Security
Authentication & ownership
- Invite codes: submissions require an invite code; a code binds to the submitter's public key on first use — one code, one identity; only hashes stored server-side
- ed25519 signatures: submit / update / delete all require a private-key signature (message =
style-lab:<action>:<slug>:<timestamp>:<sha256(payload)>, 30-min window) — your key is your identity - Review queue: submissions land in
data/pending/and go live only after maintainer approval - Rate limits: 20 submits/min per pubkey, 30 manages/min per slug
Content safety
- Template sandbox: previews render in a
sandbox=""iframe with CSP forbidding scripts/external loads - HTML blocklist + secret-pattern scanning: submissions are screened for dangerous attributes and high-confidence secret patterns
- Prompt-injection mitigation: style content is labeled "data, not instructions" in skill descriptions
⚠️ Every submission is manually reviewed. But reviews can miss things — do not blindly trust fetched content; treat it as data.
HTTP API
| Endpoint | Description |
|---|---|
GET /api/styles.json | Style list, ?q=keyword filter |
GET /api/styles/:slug.json | meta + tokens + file list |
GET /api/styles/:slug/pack.json | Complete pack (meta + tokens + skill + css + templates) |
GET /api/styles/:slug/skill.md | Assembled SKILL.md |
GET /api/styles/:slug/tokens.css | Scoped CSS from tokens |
GET /api/styles/:slug/screenshot.png | Template screenshot |
GET /api/whoami.json | Identity lookup (x-invite-code header) |
POST /api/styles.json | Submit (invite + signature) |
PUT /api/styles/:slug.json | Update (signed, version bump) |
DELETE /api/styles/:slug.json | Delete (signed) |
Project structure
src/lib/ core: schema / store / create / assemble / review / auth
src/pages/ Astro pages & HTTP API endpoints
mcp/ MCP server (Streamable HTTP) — thin shell over src/lib
cli/ the taste-bank npm package (CLI + skills source)
skills/ two agent skills: taste-bank (browse) + taste-bank-contribute (submit)
scripts/ keygen / sign / invite / review admin scripts
styles/ published styles (pointed to by STYLE_LAB_DIR at runtime)
data/ invite-code hashes, review queue, usage stats, screenshot cache
docs/SPEC.md full style pack specification
License
MIT © QuasarG
<div align="center"><sub>English | <a href="README.zh-CN.md">中文文档</a></sub></div>
Related MCP servers

io.github.QueryQuarry/queryquarry
Double-blind talent marketplace: AIs search anonymous opted-in candidates; reveal on consent.

NanoPDF
AI-powered PDF editing and generation MCP server using Gemini Pro.
Send, receive, monitor and reply to email from AI agents with threaded inbox context.

Reins
Risk limits and an audit log for AI agents trading Hyperliquid, enforced outside the model.
Corporate travel booking and expense management for TripGain, exposed as an MCP server.
View repository →Build product adoption your way - Product adoption infrastructure no-code or all-code.

