MCP Servers

How AI connects to external systems -- ADO, AEM, Figma, Chrome, and more

Concept

What is MCP?

Model Context Protocol -- a standard way for AI to call external APIs.

Direct API Access, Not Guessing

Instead of the AI guessing about your AEM instance, it queries it directly. Instead of parsing ADO or Jira web pages, it calls the tracker API through MCP. MCP is a standard protocol — any tool that implements it can be used by any AI that supports it.

Servers

KAI's MCP Servers

Six servers connecting AI to your entire development ecosystem.

ADO MCP

Work items, pull requests, repos, wiki, pipelines. Uses browser OAuth for authentication.

Tools

wit_get_work_item, repo_create_pull_request,
search_code, pipelines_run_pipeline

@azure-devops/mcp

AEM MCP

Live JCR access — components, dialogs, pages, node queries. Multi-instance support (local + QA). Included with the dx-aem plugin; also available standalone via npm.

Tools

getNodeContent, scanPageComponents,
createPage, updateComponent

aem-mcp-server

Chrome DevTools MCP

Browser automation — navigate, screenshot, click, evaluate JavaScript. Used for visual verification of AEM components.

Tools

navigate_page, take_screenshot,
click, evaluate_script

stdio

Figma MCP

Design extraction — tokens, screenshots, component context. Powers the Figma-to-code pipeline.

Tools

get_design_context, get_screenshot,
get_variable_defs

HTTP

Atlassian MCP

Work items, sprints, wiki pages. Uses personal access token (PAT) for authentication.

Tools

jira_get_issue, jira_search,
confluence_create_page, jira_add_comment

Jira + Confluence

axe MCP

Accessibility testing — WCAG violations and remediation guidance from the Deque axe engine.

Tools

analyze -- scan for violations
remediate -- suggest fixes

Docker

Flow

How It Works

Skill instructs AI, AI calls MCP, MCP queries the system, AI reasons about the result.

Example Flow

Skill tells AI “check this component” → AI calls AEM/scanPageComponents → MCP server queries AEM JCR → returns JSON → AI reasons about the result and reports findings.

Config

Configuration

Two locations: project-level and plugin-level.

Project .mcp.json

Shared by both Claude Code and Copilot CLI. For servers that apply to the whole project.

.mcp.json (ADO example)

{
"mcpServers": {
  "ado": {
    "type": "stdio",
    "command": "npx",
    "args": ["-y", "@azure-devops/mcp", "my-org"]
  }
}
}

Plugin .mcp.json

Auto-loaded when the plugin is installed. For servers bundled with a specific plugin (AEM, Chrome DevTools, Figma, axe).

Tool naming convention

# Project-level:
mcp__ado__wit_get_work_item

# Plugin-level (dx-aem plugin):
mcp__plugin_dx-aem_AEM__getNodeContent

Docs

Official Documentation

MCP Protocol

modelcontextprotocol.io

Claude Code MCP

docs.anthropic.com/…/mcp-servers

Copilot MCP

docs.github.com/…/add-mcp-servers