Skip to main content

​
Karada Documentation Guidelines & Agent Rules

​
About This Project

  • This is the official documentation site for Karada.ai, built on Mintlify.
  • Content is written in MDX files with YAML frontmatter.
  • Global navigation, SEO, and structured data configuration live in docs.json.
  • Preview locally with mint dev or check links with mint broken-links.

​
Canonical Terminology & Brand Guardrails

  1. Brand Entity Scope:
    • In all public documentation titles, OpenGraph metadata, structured schemas, and outreach copy, always write Karada.ai (to differentiate from karada.com and establish entity clarity).
    • Within internal UI text and headings, “Karada” may be used normally.
  2. Plugins (Non-Negotiable):
    • Always use Plugins / Plugins Marketplace (NEVER use “integrations”).
  3. Core Engine & Architecture:
    • Auto-MCP: Karada’s automated pipeline converting OpenAPI v3 and Swagger v2 specifications into Model Context Protocol (MCP) servers.
    • Proprietary Go Engine: Karada is a proprietary language-agnostic platform powered by a high-performance Go translation engine (NEVER refer to Karada as open-source or OSS).
    • Unified MCP Gateway: Multiplexing proxy aggregating multiple upstream MCP servers into a single Streamable HTTP / SSE endpoint.
    • Zero Fabricated Claims: Never invent benchmark multipliers or fake memory numbers.
    • No Demo Videos: Do not reference demo videos or link to video walkthroughs.

​
SEO, AEO & Agentic Discoverability Standards

  • Frontmatter Requirements: Every .mdx file must define an explicit title (clean without | Karada.ai suffix, as Mintlify handles site branding automatically), a concise sidebarTitle for the navigation table of contents, and an action-oriented description (120–160 characters).
  • AEO Definition Snippets: High-value conceptual pages must begin with a clear blockquote definition (> **What is [Feature]?**) to facilitate answer engine extraction by Perplexity, ChatGPT Search, Claude, and Gemini.
  • Diátaxis Information Architecture:
    • Tutorials: quickstart.mdx (learning-oriented).
    • How-To / Playbooks: playbooks/*.mdx (goal-oriented developer recipes for specific APIs like GitHub or Stripe).
    • Explanation: features/*.mdx (architecture-oriented deep dives).
    • Reference: troubleshooting/error-codes.mdx (information-oriented diagnostics).
  • Token Budgeting: Keep individual documentation pages under 15,000 tokens so coding agents (Cursor, Claude Code, Antigravity) can ingest complete files without lossy summarization.
  • Copy-Ready Configs: Always provide complete, copyable configuration JSON blocks for Claude Desktop (claude_desktop_config.json) and Cursor IDE settings.