AllGoodInsp MCP Server

A curated design reference database for AI agents. Retrieve structured design data from hundreds of curated websites — real CSS values, typography specs, color palettes, and design rationale — to generate better web designs.

Instead of generating from generic prompts, your AI retrieves actual design decisions from quality sites and synthesizes them into code-ready specifications.

Quick Start

Option A: OAuth (no API key needed)

On claude.ai or Claude Desktop, add a custom connector with the URL https://mcp.allgoodinsp.com and sign in with Google when prompted. Your account and free credits are provisioned automatically.

Option B: API key

Generate an API key at allgoodinsp.com/account (or use agent self-registration below), then add to your MCP configuration:

{
  "mcpServers": {
    "allgoodinsp": {
      "url": "https://mcp.allgoodinsp.com",
      "headers": {
        "Authorization": "Bearer agi_your_key_here"
      }
    }
  }
}

Configuration file locations:

If your client only supports stdio servers (no remote URL field), use the mcp-remote bridge:

{
  "mcpServers": {
    "allgoodinsp": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://mcp.allgoodinsp.com",
        "--header",
        "Authorization: Bearer agi_your_key_here"
      ]
    }
  }
}

Agent self-registration

AI agents can register without a browser and get an API key in one call:

POST https://api.allgoodinsp.com/v1/auth/register
Content-Type: application/json

{ "email": "agent@example.com", "name": "My Agent" }

The response contains an agi_ API key to use as the Bearer token above.

Related MCP server: ReftrixMCP

Basic Usage Flow

A typical design workflow with AllGoodInsp follows this sequence:

Step 1: Load the methodology

get_reference_guide()
get_methodology({ layers: ["principles", "patterns"] })

Gives your AI the design vocabulary — universal principles (typography, layout, color, motion, IA) and recurring patterns. Load once per session. This is the foundation: methodology without references produces correct but boring design.

Step 2: Search for references

search_sites({ query: "minimal SaaS landing page, confident and warm" })

Describe what you're building in natural language — mood, purpose, aesthetic. The search uses 3-axis matching (purpose + mood + contrast diversity) to return varied, relevant results.

Step 3: Select and read references

get_site({ site_id: "stripe-com", detail: "full" })
get_site({ site_id: "linear-app", detail: "full" })
get_site({ site_id: "vercel-com", detail: "full" })

Pick 3+ references. Each reference contains CSS values, typography specs, color palettes, and design rationale. Use each site as a specialist for specific decisions — don't average them.

Step 4: Synthesize into a design brief

extract_essence({
  site_ids: ["stripe-com", "linear-app", "vercel-com"],
  brief: "SaaS landing page for a developer tool, confident and minimal"
})

extract_essence combines the selected references into a code-ready design specification — CSS variables, color palette, typography scale, spacing system, section structure, and design rules with fixed/explorable boundaries.

Step 5: Build and review

Build your design using the synthesized brief, then run the quality checklist:

self_review()

Checks for design anti-patterns, craft quality, principle adherence, and IA structure.

Tools

extract_essence — Primary tool

Synthesize 2-5 site references into a code-ready design brief. Returns CSS variables, color palette, typography scale, spacing system, section structure, and design rules with fixed/explorable boundaries.

extract_essence({
  site_ids: ["stripe-com", "linear-app", "vercel-com"],
  brief: "SaaS landing page for a developer tool, confident and minimal"
})

search_sites

Semantic search across all references by mood, purpose, or description. Uses 3-axis search (purpose + mood + contrast diversity) to return varied results.

search_sites({ query: "warm editorial magazine layout" })
search_sites({ query: "bold fintech landing page", category: "software-saas" })

get_site

Retrieve the full design analysis for a specific site — color palette, typography, sections, components with CSS values and rationale.

get_site({ site_id: "stripe-com" })
get_site({ site_id: "stripe-com", detail: "full" })

search_by_component

Find sites by specific design elements. Searches across all components in all references.

search_by_component({ query: "hero with video background" })
search_by_component({ query: "pricing table with toggle" })

get_methodology

Access the design methodology — universal principles, craft guidelines, recurring patterns, and information architecture.

get_methodology({ layers: ["principles", "patterns"] })

get_principles

Design principles for specific domains.

get_principles({ category: "typography" })
get_principles({ category: "color" })

Available categories: typography, layout, color, motion, ia

get_patterns

Recurring technique combinations distilled from the collection, with evidence from multiple sites.

get_patterns()

get_reference_guide

Guide for reading and combining design references effectively.

get_workflow

Autonomous workflow for AI agents building design systems from references.

self_review

Post-implementation quality checklist for design patterns, anti-patterns, craft quality, and information architecture.

What's in the data

Each reference contains structured design decisions from a real website:

• Color palette — background, text primary/secondary, CTA, accent (with hex values and rationale)

• Typography — font families, weights, sizes, line-heights, letter-spacing for headings, body, nav, CTA

• Convention breaks — intentional violations of design principles, with scope and rationale

• Sections — hero, navigation, content, CTA, footer — each with dominant design decisions

• Components — individual elements with CSS values, principle references, and "why" explanations

Three-layer knowledge system

Categories

Agency, E-Commerce, Consulting, Software/SaaS, Portfolio, Food & Beverage, Hospitality, Education, Health & Wellness, Media/Editorial, Deep-Tech, Architecture, Business/Finance, Recruitment, and more.

REST API

For programmatic access outside MCP, a REST API is also available.

Base URL: https://api.allgoodinsp.com/v1

Search responses include a next_step field pointing at /extract-essence with an example body. Generate an API key at allgoodinsp.com/account or via POST /auth/register. See the API docs for parameters, credits, and rate limits.

Links

• Gallery: allgoodinsp.com

• MCP Endpoint: https://mcp.allgoodinsp.com

• REST API: https://api.allgoodinsp.com/v1

• MCP Docs: allgoodinsp.com/docs/mcp

• API Docs: allgoodinsp.com/docs/api

License

MIT