Angular CLI MCP Server

The Angular CLI includes a Model Context Protocol (MCP) server that enables AI assistants (like Cursor, Antigravity, JetBrains AI, etc.) to interact directly with the Angular CLI. It provides tools for code generation, workspace analysis, and running builds/tests.

Get Started

To use the MCP server, you configure your host environment (IDE or CLI) to run npx @angular/cli mcp.

Create a file named .antigravity/mcp.json in your project's root:

json

{
  "mcpServers": {
    "angular-cli": {
      "command": "npx",
      "args": ["-y", "@angular/cli", "mcp"]
    }
  }
}

Create .cursor/mcp.json in the project root (or globally at ~/.cursor/mcp.json):

json

{
  "mcpServers": {
    "angular-cli": {
      "command": "npx",
      "args": ["-y", "@angular/cli", "mcp"]
    }
  }
}

Create .vscode/mcp.json:

json

{
  "servers": {
    "angular-cli": {
      "command": "npx",
      "args": ["-y", "@angular/cli", "mcp"]
    }
  }
}

Available Tools (Default)

When the MCP server is enabled, AI agents have access to the following tools:

Name	Description
`ai_tutor`	Launches an interactive AI-powered Angular tutor.
`devserver.start`	Asynchronously starts a dev server (`ng serve`). Returns immediately.
`devserver.stop`	Stops the dev server.
`devserver.waitforbuild`	Returns the logs of the most recent build in a running dev server.
`getbestpractices`	Retrieves the Angular Best Practices Guide (crucial for standalone components, typed forms, etc.).
`list_projects`	Lists all applications and libraries in the workspace by reading `angular.json`.
`onpushzonelessmigration`	Analyzes code and provides a plan to migrate it to `OnPush` change detection (prerequisite for zoneless).
`run_target`	Executes a configured target (e.g., build, test, lint, e2e, deploy).
`search_documentation`	Searches the official documentation at `https://angular.dev`.

Common Workflows

These workflows demonstrate how AI assistants coordinate different MCP tools to automatically achieve complex developer stories.

1. Performance Tuning: Zoneless & OnPush Migration

The AI agent optimizes change detection performance and migrates components to a zoneless-ready state.

Discover Workspace: The AI agent calls list_projects to locate components, projects, and style/test configurations in the workspace.
Schematic Modernization (Prerequisite): The AI agent runs any prerequisite signal migrations using standard ng generate commands (e.g., Signal Inputs, Signal Queries).
Plan Migration: The AI agent calls onpushzonelessmigration with the absolute path of the directory or component file.
Apply Changes: The AI agent automatically applies the single actionable change returned by the tool to the codebase.
Verify Changes: The AI agent runs unit tests by calling run_target with the target parameter set to "test".
Repeat: The AI agent calls onpushzonelessmigration again to retrieve the next step, repeating until the tool indicates the migration is complete.

2. Feature Development & TDD Loop

The AI agent automates research, implementation, and verification when developing new features.

API & Syntax Research: The AI agent uses search_documentation to look up Angular APIs or syntax rules (e.g., @defer block options).
Load Coding Standards: The AI agent calls getbestpractices with the workspace path to load Angular version-aligned coding rules.
Start Local Dev Server: The AI agent starts a background server by calling devserver.start.
Monitor Build: The AI agent uses devserver.waitforbuild to watch build logs and ensure compilation succeeds as it edits the code.
Write and Execute Tests: The AI agent identifies the project's test framework (e.g., Jasmine, Jest, Vitest) via listprojects, writes the corresponding test file, and runs the tests using runtarget with "test".
Stop Dev Server: When finished, the AI agent stops the active dev server by calling devserver.stop.

3. Developer Onboarding and Learning

The AI agent guides the developer through Angular concepts in an interactive sandbox.

Discover Projects: The AI agent calls list_projects to scan the workspace and identify the codebase structure.
Launch Tutor: The AI agent runs ai_tutor to load the curriculum instructions, persona, and tutoring guidelines.
Follow the Curriculum: The AI agent guides the user through the curriculum, explaining concepts and instructing them on what components to build or modify.
Implement & Verify: The AI agent helps implement the sandbox code and verifies changes using run_target with "test" or "build".

Command Options

You can pass arguments to the MCP server in the args array of your configuration:

--read-only: Only registers tools that do not modify the project.
--local-only: Only registers tools that do not require an internet connection.

Example for read-only mode:

json

"args": ["-y", "@angular/cli", "mcp", "--read-only"]