Use the Microsoft Learn CLI

The Microsoft Learn CLI (mslearn) provides terminal-friendly access to the public Microsoft Learn Model Context Protocol (MCP) Server. You can use it to search Microsoft documentation, fetch Learn articles, search official code samples, and diagnose your connection.

By default, the CLI connects to the following endpoint:

https://learn.microsoft.com/api/mcp

Prerequisites

The Microsoft Learn CLI requires Node.js 22 or later. To check your installed version, run:

node --version

Install the CLI

You can run the CLI without installing it:

npx @microsoft/learn-cli search "azure functions timeout"

Or install it globally:

npm install -g @microsoft/learn-cli
mslearn search "azure functions timeout"

The remaining examples use the mslearn command, which is available after a global installation. If you run the CLI without installing it, replace mslearn with npx @microsoft/learn-cli.

Installing the npm package doesn't configure agent discovery. To make the CLI available as a fallback for supported coding agents, complete the steps in Configure agent discovery.

Configure agent discovery

The standalone CLI can install a CLI-first skill for GitHub Copilot, Claude Code, and Codex. The skill teaches the agent to run npx @microsoft/learn-cli@latest when Microsoft Learn MCP tools aren't available.

To install discovery for supported agents detected in your user profile, run:

mslearn setup --cli

To detect agents in the current repository and install discovery at the project level, run:

mslearn setup --cli --project

You can override automatic detection by specifying one or more agents:

mslearn setup --cli --copilot
mslearn setup --cli --claude
mslearn setup --cli --codex
mslearn setup --cli --copilot --claude --codex

Add --project to any of these commands to install discovery only in the current repository:

mslearn setup --cli --copilot --project
mslearn setup --cli --claude --project
mslearn setup --cli --codex --project

The setup command creates the following skill files:

Agent Scope Skill file
GitHub Copilot User ~/.copilot/skills/microsoft-learn-cli/SKILL.md
GitHub Copilot Project .github/skills/microsoft-learn-cli/SKILL.md
Claude Code User ~/.claude/skills/microsoft-learn-cli/SKILL.md
Claude Code Project .claude/skills/microsoft-learn-cli/SKILL.md
Codex User ~/.agents/skills/microsoft-learn-cli/SKILL.md
Codex Project .agents/skills/microsoft-learn-cli/SKILL.md

The --cli option is required. If you don't specify an agent, setup detects installed agents from their known user or project directories. Explicit targets (--copilot, --claude, and --codex) override detection and can be combined. Agents outside these plugin ecosystems, including Cursor, aren't installed by this workflow.

Re-running setup refreshes content managed by the Microsoft Learn CLI.

Remove agent discovery

To remove managed discovery files without changing unrelated agent configuration, run:

mslearn remove --cli
mslearn remove --cli --copilot
mslearn remove --cli --claude --codex --project

Without an explicit agent target, the remove command detects only discovery artifacts managed by the Microsoft Learn CLI.

Compare CLI-first and MCP-first integrations

Mode Installation Agent behavior
Standalone CLI-first Install or run the npm package, and then use mslearn setup --cli with an optional agent target. Detected or selected agents invoke the standalone CLI through npx when Learn MCP tools aren't available. No MCP configuration is added.
Repository plugin, MCP-first Install the repository plugin for GitHub Copilot, Claude Code, or Codex. The plugin supplies the Microsoft Learn MCP endpoint and MCP-oriented skills.

When both integrations are installed, agents should prefer the Microsoft Learn MCP tools and use the standalone CLI skill only as a fallback.

Use CLI commands

The following examples show the available commands:

mslearn search "azure functions timeout"
mslearn fetch "https://learn.microsoft.com/azure/azure-functions/functions-versions"
mslearn fetch "https://learn.microsoft.com/azure/azure-functions/functions-versions" --section "Function app timeout duration"
mslearn fetch "https://learn.microsoft.com/azure/azure-functions/functions-versions" --max-chars 3000
mslearn code-search "cosmos db change feed processor"
mslearn code-search "cosmos db change feed processor" --language csharp
mslearn doctor
mslearn doctor --format json

The CLI supports these commands and options:

  • search <query> [--json] searches official Microsoft documentation.
  • fetch <url> fetches a Learn page as Markdown-friendly output.
  • fetch <url> --section <heading> returns a single section.
  • fetch <url> --max-chars <number> truncates the output.
  • code-search <query> --language <name> [--json] searches official code samples.
  • doctor [--format text|json] checks the runtime and MCP server connectivity.
  • setup --cli [--copilot|--claude|--codex] [--project] detects agents or installs discovery for explicit targets.
  • remove --cli [--copilot|--claude|--codex] [--project] detects and removes only managed discovery content.

The search and code-search commands return human-readable text by default. Use --json to return the raw JSON response for processing with other command-line tools:

mslearn search "azure functions" --json | jq ".results[].title"
mslearn code-search "BlobServiceClient" --language python --json

Configure the endpoint

To use a different endpoint, set the MSLEARN_ENDPOINT environment variable or pass --endpoint <url> for an individual command.

For example, in PowerShell:

$env:MSLEARN_ENDPOINT = "https://learn.microsoft.com/api/mcp"
mslearn doctor

Next steps

To learn more about the Microsoft Learn MCP Server, visit the following articles:

  • Last updated on