Taste Bank MCP Server

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.

Related MCP server: Wedges

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 setup installs the CLI and injects skills into all your coding agents (Claude Code, Codex, Cursor, ZCode, Kimi Code, and 70+ more via the open skills ecosystem)

• 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)

Contributing (invite + ed25519 identity)

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 styles

• taste-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

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