AI Agent Integration (MCP)

Last reviewed: August 2026

The CloudPick documentation site provides a Model Context Protocol (MCP) server. Connect it to your AI coding agent (Kiro, Claude Desktop, Cursor, etc.) to search and reference documentation during conversations.

Endpoint

https://docs.cloudpick.kr/mcp

Protocol: MCP Streamable HTTP (POST)

This is the same origin as the docs site. /mcp is the MCP server.

Legacy GitBook endpoint

The old GitBook address https://docs.cloudpick.kr/~gitbook/mcp still connects, but it’s now a notice-only stub — it no longer serves the real documentation tools. tools/list returns a single migration_notice tool instead of list_docs/search_docs/get_doc, and calling any tool name just returns text telling you to update your configuration to use /mcp. The same note is also added to the initialize response’s instructions. Search and lookup no longer work through this address — update your configuration to /mcp.

Client Configuration

Kiro

.kiro/settings/mcp.json (workspace) or ~/.kiro/settings/mcp.json (global):

{
  "mcpServers": {
    "cloudpick-docs": {
      "type": "streamable-http",
      "url": "https://docs.cloudpick.kr/mcp"
    }
  }
}

Claude Desktop

claude_desktop_config.json:

{
  "mcpServers": {
    "cloudpick-docs": {
      "type": "streamable-http",
      "url": "https://docs.cloudpick.kr/mcp"
    }
  }
}

Cursor

Settings → MCP Servers → Add:

{
  "cloudpick-docs": {
    "url": "https://docs.cloudpick.kr/mcp"
  }
}

VS Code (GitHub Copilot)

.vscode/mcp.json:

{
  "servers": {
    "cloudpick-docs": {
      "type": "http",
      "url": "https://docs.cloudpick.kr/mcp"
    }
  }
}

Available Tools

Tool	Description	Parameters
list_docs	Returns the full list of document titles	lang (optional)
search_docs	Search docs by keyword (ko·en·ja)	query, limit (default 5), lang (optional)
get_doc	Returns full markdown of a single document	title (partial match supported), lang (optional)

Multilingual Support

CloudPick docs are available in Korean (ko), English (en), and Japanese (ja), and the MCP server auto-detects the request language to return docs in that language.

• One endpoint URL, language-independent (https://docs.cloudpick.kr/mcp). You do not need a different URL per language.
• The default language is Korean (ko). Korean is the source of truth (SOT); when there is no language signal, the server always responds in ko.
• Auto-detection: the language is inferred from the query/title script (Hangul → ko, Kana → ja). Latin-only queries (e.g., EKS, S3) carry no language signal and fall back to ko.
• Explicit selection: to be certain, pass the optional lang parameter (ko·en·ja) to any tool. Setting it to the user’s conversation language is recommended.
• Language header: every response starts with a [lang=ko | also available: en, ja | source_lang=ko | to get another language, call again with lang="en"] header. It tells you which language was returned, which other versions exist, and how to re-request another language, so the agent can call again with lang when needed.

Language resolution — the same for all three tools

Language is resolved in this order: ① explicit lang parameter → ② script detection of the query/title (Hangul→ko, Kana→ja) → ③ default ko. All three tools (list_docs·search_docs·get_doc) use this identical rule — there are no per-tool exceptions.

This means Latin-only queries/titles (e.g., EKS, CDN, API Gateway) cannot be identified and respond in ko. To get English or Japanese, specify lang explicitly as shown below.

Getting answers in English or Japanese

Because Latin-only queries and titles respond in Korean (the default), specify the lang parameter explicitly to receive English or Japanese responses. Most MCP clients let you pass this via the tool arguments — instruct your agent to pass the user’s conversation language as lang.

search_docs({ query: "EKS", lang: "en" })          // search in English
get_doc({ title: "VPC and Subnets", lang: "en" })   // fetch the English doc
get_doc({ title: "VPCとサブネット", lang: "ja" })    // fetch the Japanese doc

Queries in Hangul or Kana are auto-detected and respond in ko or ja respectively without lang. If a document does not exist in the detected language, another language version is located and returned (cross-language lookup); in that case the lang= value in the response header tells you the actual returned language.

Usage Examples

You can ask your AI agent things like:

• “Search CloudPick docs for serverless best practices”
• “Look up the CSAP certification tier system in CloudPick”
• “Show me the full VPC and Subnets document from CloudPick”

The agent uses search_docs to find relevant documents, then get_doc to read the full content and incorporate it into its response.

Related Documents

• Introduction
• Glossary