Human Taste Engine
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Human Taste EngineCapture example.com and critique its homepage design"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Don’t stop at “it works.”
Give your AI a sharper creative workflow: understand the brief, make work with intent, inspect the real result, and improve the biggest problem before calling it done.
Human Taste Engine (HTE) is a portable, AI-agnostic skill for creative work. It guides an AI through understanding, research, creation, critique and verification—while your AI remains the reasoning and execution engine.
⬇ Download the portable skill (v1.0.2) · See all releases · Browse the skill
A skill, not an app. The download contains Markdown guidance for your AI. Human Taste Engine does not need its own program, runtime, server, API key or installed tools.
How it works
Instead of treating the first output as finished, HTE guides the host AI through a focused loop:
flowchart LR
A[Understand the brief] --> B[Choose a fitting direction]
B --> C[Make the work]
C --> D[Inspect what can actually be checked]
D --> E[Critique and improve the biggest issue]
E --> F[Deliver with limits stated honestly]The aim is more considered work with fewer avoidable correction rounds—not a forced visual style or a promise that every AI will follow every instruction.
Related MCP server: RenderProof
Try it in three steps
Download the portable skill ZIP and extract it.
Give your AI the extracted
human-taste-engine/folder. Install it through the host’s skill support, or provide the folder to an AI that can read Markdown instructions. In the source repository, this same folder lives atskills/human-taste-engine/.Ask for the work you want. For example:
Create a landing page for my independent clothing brand. It should feel confident and crafted, not like a generic luxury template. Use Human Taste Engine, make a fitting direction, and inspect what you can actually verify before you finish.
If the host has no skill installer, provide the folder and say:
Read
SKILL.mdand follow it for this task. Open only the supporting guidance it routes to. Use capabilities you already have, and tell me what you could not verify.
What changes in the workflow
Without a deliberate review loop | With Human Taste Engine |
Request → guess → generate → done | Understand → choose a direction → create → inspect → critique → refine |
“Premium” becomes a stock visual preset | Style choices depend on the project, audience and references |
A successful build stands in for quality | The AI checks the real output when its tools allow it—and labels unchecked judgments honestly |
Every task gets the same amount of process | Small edits stay small; substantial creative work gets a bounded review loop |
HTE asks the host to separate what is known, observed, inferred and unknown; use research when fresh evidence matters; preserve the user’s intent; and fix the highest-impact supported issue first. Anti-patterns are prompts to investigate, not automatic bans.
Where it helps
Websites and interfaces
Design direction, layout and typography
Image and video work
Presentations
Marketing and product copy
Creative reviews, comparisons and redesigns
For a typo or other mechanical edit, the skill keeps the process lightweight. For a major visual task, it routes the AI to only the relevant guidance and asks it to inspect available evidence before delivery.
Works with your AI’s existing capabilities
The core skill uses tools the host already has—such as a browser, visual input, search, files or media tools. It does not provide those tools or install them. If the host cannot render a page, view an image, play a video or check a live fact, HTE asks it to say that the judgment remains unverified.
The skill is written in the open Agent Skills format. Use it with an AI host that supports Agent Skills, or give the folder to a host that can read supplied Markdown files. The exact installation method depends on the AI host. An optional Claude Code marketplace adapter points to this same portable skill; it is not required.
What’s in the repository
skills/human-taste-engine/— the skill and its focused supporting guidance.examples/— examples of how the method can be applied.evaluations/— manual behavior cases, not executable tests or guarantees of host compliance.THIRD_PARTY_NOTICES.md— source and license notices.
Limits
Human Taste Engine is guidance for an AI host. It cannot force a host to comply, add capabilities the host lacks or guarantee a particular creative result. It does not claim visual, motion, playback or interaction checks unless the host actually performed them.
Ready to try it? Download the skill ZIP, give the folder to a compatible AI, and ask it to use Human Taste Engine.
Licensed under MIT.
This server cannot be deployed
Maintenance
Related MCP Connectors
Evidence-readiness MCP server: validate, audit, and score briefs, memos, and evidence packs.
MCP server for visual regression testing: triage a PR's UI diffs from your coding agent.
MCP-native web evidence and claim verification: cited, source-grounded evidence for AI agents.
MCP server providing access to the Scorecard API to evaluate and optimize LLM systems.
Related MCP Servers
- AlicenseNot gradedqualityCmaintenanceA read-only MCP server that gives AI agents the web as compact, ranked, verified evidence — no API keys, no cloud retrieval, all models local.64 npm1MIT
- AlicenseAqualityDmaintenanceA local-first MCP server that captures rendered web evidence including screenshots, motion, pixel diffs, and CSS animation metadata for coding agents.71MIT
- AlicenseAqualityBmaintenanceA local-first MCP server for retrieving a small evidence set and recording reviewed conclusions, policy-gated and redacted without giving an agent general filesystem access.5MIT
- AlicenseAqualityBmaintenanceA local-first, model-neutral MCP server for collecting and normalizing change-scoped release evidence. It provides deterministic Git change summaries, evidence collection, and review bundles for agent review.78 npmApache 2.0