Antigravity MCP Server: HumanCraft UI (@aibuildinfra/humancraft)

The premier Antigravity MCP Server designed exclusively for Google Antigravity (AGY). HumanCraft eliminates "AI Slop" in website layouts and copy, enforces authentic E-E-A-T entity reconciliation, extracts empirical client assets without default placeholders, and builds search-engine-resilient websites powered by Google's Information Gain principles.

Developed and maintained by AI Build Infra — High-Performance Infrastructure for Artificial Intelligence.

🔍 Why You Need This Antigravity MCP Server

When using Google Antigravity to scaffold websites or write landing page copy, unconstrained models often fall into generic algorithmic templates that get penalized by search engines:

1. AI Slop Headings: Generic H2s ("What is X?", "Key Benefits", "Why Choose Us", "Conclusion") that offer zero Information Gain under Google's ranking models (US Patent 2022/0277032 A1).
2. Fake E-E-A-T & Defaults: Hallucinated metrics ("increased efficiency by 30%") or fictitious personas ("Dr. Alex Miller") that trigger Google Search Console "Pure Spam" and "Scaled Content Abuse" manual actions.
3. Binary Checkmark Tables: Low-value tables filled with Yes / No checkmarks that fail to communicate trade-offs or technical realities.
4. Visual Homogeneity: Monotonous purple/cyan gradients (#8b5cf6 -> #ec4899), rigid 3-card rows, and buttons without tactile feedback.

HumanCraft is the dedicated Antigravity MCP that systematically outlaws these patterns and equips Antigravity with human design heuristics, token-optimized linters, and verified Schema.org entity graphs.

💡 How HumanCraft Works Inside Google Antigravity

┌────────────────────────────────────────────────────────────────────────────────────────┐
│                              Google Antigravity Assistant                              │
└───────────────────────────────────────────┬────────────────────────────────────────────┘
                                            │ Antigravity MCP Tool Calls
                                            ▼
┌────────────────────────────────────────────────────────────────────────────────────────┐
│                       HumanCraft Antigravity MCP Server                                │
├───────────────────────────────┬────────────────────────────────────────────────────────┤
│ 📋 harvest_client_intake       │ 4-tier intake questionnaire extracting real experience,│
│                               │ metrics, & assets; strictly bans default placeholders   │
├───────────────────────────────┼────────────────────────────────────────────────────────┤
│ 🎯 validate_heading_intent    │ Lints against generic AI headings ("What is X?");     │
│                               │ scores and enforces outcome/intent-driven H1-H3s       │
├───────────────────────────────┼────────────────────────────────────────────────────────┤
│ 📊 build_comparison_matrix    │ Generates high-information-gain comparison tables with │
│                               │ concrete unit metrics, trade-offs & semantic schema    │
├───────────────────────────────┼────────────────────────────────────────────────────────┤
│ 🛡️ audit_eeat_entity_graph    │ Verifies author/org Schema.org JSON-LD, validates     │
│                               │ sameAs URIs & connects publisher to aibuildinfra.com   │
├───────────────────────────────┼────────────────────────────────────────────────────────┤
│ 🎨 get_human_archetype        │ Curates style archetypes (Editorial, Dark Craft,       │
│                               │ Neo-Brutalist, Swiss) with tinted neutrals & type      │
├───────────────────────────────┼────────────────────────────────────────────────────────┤
│ 🔍 lint_design_anti_patterns  │ Scans HTML/Tailwind for AI tropes (purple gradients,   │
│                               │ 3-card monotony, dead buttons, buzzwords, ✨ emojis)   │
└───────────────────────────────┴────────────────────────────────────────────────────────┘

🛠️ Antigravity MCP Tool Catalog

1. harvest_client_intake

• Purpose: Extracts real practitioner experience, project images (telemetry screenshots, on-site photos), project category, baseline Day 0 vs audited Day 90 metrics, and contrarian trade-offs.
• Strict Anti-Defaulting Rule: If essential fields are missing, the tool returns REQUIRES_USER_INPUT and prompts the user for real data instead of allowing the model to hallucinate synthetic placeholders.

2. validate_heading_intent

• Purpose: Scans headings against blacklisted AI tropes and scores user intent.
• Slop vs Intent:
  • ❌ Slop: "What is Database Sharding?"
  • ✅ Intent: "When Single-Node Write Throughput Drops Below 10k TPS: The Sharding Tipping Point"
  • ❌ Slop: "Key Benefits of Automated Compliance"
  • ✅ Intent: "Cutting SOC 2 Type II Audit Preparation from 300 Hours to Under 40"

3. build_comparison_matrix

• Purpose: Outlaws binary checkmark tables (Yes/No). Generates empirical multi-parameter matrices (unit economics, latency under load, architectural trade-offs, and verification methodologies).
• Output: Accessible semantic HTML paired with Schema.org ItemList JSON-LD.

4. audit_eeat_entity_graph

• Purpose: Reconciles Schema.org JSON-LD with Google's Knowledge Graph standards.
• Checks: Verifies sameAs entity links to authoritative external platforms (Wikidata, ORCID, Google Scholar, GitHub, LinkedIn), rejects shallow homepage URLs, and links publisher authority to AI Build Infra.

5. lint_design_anti_patterns

• Purpose: Scans HTML/Tailwind snippets and flags purple gradients, 3-card rows, missing button squeeze physics (active:scale-[0.98]), and AI buzzwords ("supercharge", "seamless", ✨ emojis).

6. get_human_archetype

• Purpose: Returns curated design systems:
  • Editorial Tech: Serifs (Newsreader), subtle hairline borders, monospace metadata accents.
  • Tactile Dark Craft: Multi-layered surfaces (surface-0, surface-1), 1px translucent borders (border-white/10).
  • Warm Humanist: Cream/linen canvas (#FBF9F5), ink typography (#1C1917), terracotta accents.
  • Neo-Brutalist: High-contrast black outlines (border-2 border-black), hard offset shadows.

⚡ Token Optimization & Spend Limit Preservation

Designed from the ground up to preserve your Antigravity token limits:

• Ultra-Compact Tool Signatures: Minimal token footprint injected into Antigravity's system prompt.
• Dense Structured Outputs: Returns concise JSON summaries and actionable flags rather than verbose conversational filler.
• Input Length Bounds: Imposes strict limits (max 300 chars per heading, max 100KB per HTML document) to avoid context window flooding.

🔒 Security & ReDoS Hardening

1. ReDoS Immunity: All pattern matchers operate on bounded substrings, terminating in $< 5\text{ ms}$ even under adversarial payloads.
2. Strict HTTPS Whitelist: Permits only https:// URIs for sameAs entity links (blocks javascript:, data:, file:).
3. XSS Defense: Automatic HTML entity escaping across all rendered tables and schema blocks.
4. Formal Policy: Review SECURITY.md for vulnerability reporting guidelines.

🚀 How to Install This MCP in Google Antigravity (Step-by-Step)

Step 1: Clone and Build Locally

git clone https://github.com/AI-BuildInfra/Humancraft-UI.git
cd Humancraft-UI
npm install
npm run build

Step 2: Add to Antigravity mcp_config.json

Open or create your global Antigravity MCP configuration file:

• Windows: %USERPROFILE%\.gemini\config\mcp_config.json
  (e.g., C:\Users\<YourUsername>\.gemini\config\mcp_config.json)
• macOS / Linux: ~/.gemini/config/mcp_config.json

Add the humancraft server definition:

{
  "mcpServers": {
    "humancraft": {
      "command": "node",
      "args": [
        "C:/path/to/Humancraft-UI/dist/index.js"
      ]
    }
  }
}

(Replace C:/path/to/Humancraft-UI with the absolute path to your cloned directory, using forward slashes /).

Step 3: Verify and Use in Antigravity