Microsoft Learn
Microsoft's own documentation server for Azure, .NET, Windows and M365. Three read-only tools — docs search, code-sample search and page fetch — no sign-in. It touches no Azure subscription and no Microsoft account, verified live 2026-08-18.
Verified connector
Listed by Anthropic as a partner connector in its Connectors Directory.
Connection checked by Agentman on .
Anthropic states this reflects the level of review a connector received, not a security audit.
Connect Microsoft Learn via MCP
https://learn.microsoft.com/api/mcpWorks in any MCP-compatible client. In Agentman Studio it is one click — no config file to edit.
Microsoft Learn Tools & Capabilities (3)
Search official Microsoft/Azure documentation to find the most relevant and trustworthy content for a user's query. This tool returns up to 10 high-quality content chunks (each max 500 tokens), extracted from Microsoft Learn and other official sources. Each result includes the article title, URL, and a self-contained content excerpt optimized for fast retrieval and reasoning. Always use this tool to quickly ground your answers in accurate, first-party Microsoft/Azure knowledge. ## Follow-up Pattern To ensure completeness, use microsoft_docs_fetch when high-value pages are identified by search. The fetch tool complements search by providing the full detail. This is a required step for comprehensive results.
Search for code snippets and examples in official Microsoft Learn documentation. This tool retrieves relevant code samples from Microsoft documentation pages providing developers with practical implementation examples and best practices for Microsoft/Azure products and services related coding tasks. This tool will help you use the **LATEST OFFICIAL** code snippets to empower coding capabilities. ## When to Use This Tool - When you are going to provide sample Microsoft/Azure related code snippets in your answers. - When you are **generating any Microsoft/Azure related code**. ## Usage Pattern Input a descriptive query, or SDK/class/method name to retrieve related code samples. The optional parameter `language` can help to filter results. Eligible values for `language` parameter include: csharp javascript typescript python powershell azurecli al sql java kusto cpp go rust ruby php
Fetch and convert a Microsoft Learn documentation webpage to markdown format. This tool retrieves the latest complete content of Microsoft documentation webpages including Azure, .NET, Microsoft 365, and other Microsoft technologies. ## When to Use This Tool - When search results provide incomplete information or truncated content - When you need complete step-by-step procedures or tutorials - When you need troubleshooting sections, prerequisites, or detailed explanations - When search results reference a specific page that seems highly relevant - For comprehensive guides that require full context ## Usage Pattern Use this tool AFTER microsoft_docs_search when you identify specific high-value pages that need complete content. The search tool gives you an overview; this tool gives you the complete picture. ## URL Requirements - The URL must be a valid HTML documentation webpage from the microsoft.com domain - Binary files (PDF, DOCX, images, etc.) are not supported ## Output Format markdown with headings, code blocks, tables, and links preserved.
Read from the server on 2026-08-18, including each tool's own safety annotations.
Limits
- It retrieves documentation; it does not operate anything. No tool on the live surface reaches an Azure subscription, a Microsoft 365 tenant, a Learn profile or any private content. Every command it drafts still has to be run by you through some other route.
- Prompts and resources are declared but empty. The server's
initializeresponse declares four capabilities —logging,prompts,resourcesandtools. Callingprompts/list,resources/listandresources/templates/liston 2026-08-18 returned empty arrays from all three. That is drift, not absence: an undeclared capability returns JSON-RPC error −32601, and none of these did. A client that renders a prompt picker for this server will show an empty menu. - Rate limits exist but are unpublished. Microsoft's FAQ confirms limits are in place and gives no number, window, or retry guidance. Plan for throttling you cannot predict.
- No filtering, and the FAQ says so. Microsoft states the server provides no filtering capability, and that scoping has to be expressed inside your question. The only structured filter on the surface is the optional
languageparameter on code-sample search. - Fetch is uncapped.
maxTokenBudgettruncates search responses only; Microsoft states fetch always returns the full page. A long Azure reference article is a large, unbounded context injection. - The primary search tool declares no required parameter, and fetch declares no output schema. Both asymmetries are documented above. Neither is a defect; both change how a strict client must behave.
- Microsoft asks you not to treat the endpoint as an API. Its developer reference states that although the endpoint is public it is not an API in the traditional sense, that the interface may change dynamically including tools and their request and response formats, and that developers should integrate through an agent framework rather than calling the endpoint directly. It declines to document request and response details for that reason. Anything on this page derived from a schema is a snapshot of 2026-08-18, not a contract.
- Browsers get a 405. We confirmed a plain GET returns HTTP 405 Method Not Allowed. That is expected, documented and not an outage.
- The OpenAI-compatible endpoint is a different surface with stale descriptions. See the note below. It is experimental, and Microsoft says experimental features are subject to change.
- The Terms of Use are the general Microsoft Learn terms. Microsoft's overview page states that using the server means agreeing to them, and those terms carry a personal and non-commercial use limitation on content obtained from the Services. The MCP FAQ separately says you may use the server to power your own agents and build agents for others, under those same terms. Read both before building a commercial product on it; we take no position on how they reconcile.
- No SLA statement was found for this free public endpoint, and the repository's Creative Commons Attribution 4.0 licence governs the published content rather than the hosted service.
- We did not call any tool. Our check was a read-only
initialize,tools/list,prompts/list,resources/listandresources/templates/listhandshake. Every behavioural claim above comes from the server's own schemas and descriptions, or from Microsoft's published documentation.
Frequently asked questions
Can the Microsoft Learn MCP server access my Azure subscription?
No. We read all three live tool schemas on 2026-08-18 and the complete set of inputs across them is a search query, an optional programming language, and a documentation URL. There is no subscription ID, tenant, resource group, credential or account field anywhere. Microsoft states the server carries publicly available documentation, not training or user profile information.
Do you need a Microsoft account to use the Microsoft Learn MCP server?
No. Microsoft's overview page states plainly that no authentication is required and that the server is publicly available at no charge. Our anonymous handshake on 2026-08-18 returned all three tool definitions with no credential attached, and Anthropic's directory records the connector as authless. The only prerequisites are an MCP client and network access.
What is the difference between microsoft_docs_search and microsoft_code_sample_search?
Search returns prose passages; code-sample search returns runnable snippets. The docs tool returns up to ten content chunks capped at 500 tokens each, aimed at grounding an explanation. The code tool returns snippet, language and source link, and takes an optional language filter across fifteen values. Microsoft's own guidance is to use the code tool whenever you are generating Microsoft or Azure code.
Why must microsoft_docs_fetch be called after a search?
Because it takes a URL it cannot invent. Fetch requires a valid HTML page on the microsoft.com domain, and the reliable source of such a URL is a search result. Microsoft describes the ordering directly: search gives breadth, fetch gives depth. Nothing in the protocol enforces the sequence, so a wrong or guessed URL simply fails rather than being rejected upfront.
Are the Microsoft Learn MCP tools safe to auto-approve?
All three are annotated readOnlyHint true, destructiveHint false and idempotentHint true, read live on 2026-08-18. None accepts a credential or an account identifier, so there is nothing to spend, change or leak. The residual risk is context poisoning rather than state change: the tools inject fetched text into your model's window, and that text is Microsoft's rather than yours.
What content does the Microsoft Learn MCP server actually cover?
Microsoft's published documentation corpus: Azure, .NET, Windows, Microsoft 365 and the rest of learn.microsoft.com, plus code samples embedded in those pages. Microsoft states the underlying knowledge service refreshes incrementally after content updates and performs a full refresh once a day. It carries no training records, no user profile data and no non-Microsoft sources.
Does the Microsoft Learn MCP server have rate limits?
Yes, though Microsoft publishes no numbers. Its FAQ states that as a public and free endpoint the service has rate limits in place to ensure fair usage, and asks for responsible use. No quota, window or retry guidance is documented, and no rate-limit headers appeared in our handshake responses on 2026-08-18. Treat heavy automated looping as unsupported.
Sources
- Microsoft Learn, "Microsoft Learn MCP Server overview" (retrieved 2026-08-18, HTTP 200; page metadata gives
ms.date2026-05-05 andupdated_at2026-05-22). This is the exactdocumentationvalue in Anthropic's directory entry, and it is current — no redirect, no migration. Source for the endpoint, the no-authentication and free-and-publicly-available statements, the 405 note, the Limitations section including the documentation-not-training-or-profile boundary and the daily refresh cadence, and the Terms of Use agreement. · retrieved 2026-08-18 - Microsoft Learn, "Microsoft Learn MCP Server developer reference documentation" (retrieved 2026-08-18, HTTP 200;
ms.date2025-11-28). The three tool names, the standard JSON client config, and the statement that the endpoint is not an API in the traditional sense and that tools and formats may change dynamically. · retrieved 2026-08-18 - Microsoft Learn, "Best practices for using the Microsoft Learn MCP Server" (retrieved 2026-08-18, HTTP 200;
ms.date2026-05-05). ThemaxTokenBudgetparameter, the statement that it affects search only because fetch always returns the full page, and the dynamic tool-discovery guidance. · retrieved 2026-08-18 - Microsoft Learn, "Microsoft Learn MCP Server frequently asked questions" (retrieved 2026-08-18, HTTP 200;
ms.date2026-05-05). The rate-limit statement, the no-filtering statement, the daily data-change statement, and the permission to power your own agents under the Terms of Use. · retrieved 2026-08-18 - Microsoft Learn, "Microsoft Learn MCP Server release notes" (retrieved 2026-08-18, HTTP 200;
ms.date2026-05-05). Server released 12 June 2025, fetch tool 6 August 2025, code-sample search 24 September 2025, general availability 7 November 2025, OpenAI-compatible endpoint 10 December 2025, CLI 10 March 2026, plugin 23 March 2026. · retrieved 2026-08-18 - Microsoft Learn, "Get started with the Microsoft Learn MCP Server" (retrieved 2026-08-18, HTTP 200;
ms.date2026-02-11). Prerequisites, the recommended instructions file, and the/plugin marketplace add microsoftdocs/mcpinstall route that disagrees with the README. · retrieved 2026-08-18 MicrosoftDocs/mcprepository (README,.mcp.json,.claude-plugin/marketplace.json,LICENSEandskills/microsoft-code-reference/SKILL.mdall retrieved 2026-08-18, HTTP 200). Microsoft's redirectaka.ms/learnmcpdocs/reporesolves here. Source for the per-client install table, the troubleshooting table, the three Agent Skills, the CLI, the experimental features, and the Creative Commons Attribution 4.0 licence. Two corroborations worth naming:.mcp.jsoncarries the endpoint byte-for-byte identical to the directory'sremote.url, andmarketplace.jsondeclares the marketplace asmicrosoft-docs-marketplace, contradicting the README'sclaude-plugins-official. · retrieved 2026-08-18- Microsoft Learn Terms of Use (retrieved 2026-08-18, HTTP 200;
ms.date2025-05-12). Consulted for the personal and non-commercial use limitation noted in Limits. It contains no clause specific to MCP or AI assistants. · retrieved 2026-08-18 - Microsoft Privacy Statement — , the
privacy_policyvalue in Anthropic's directory entry. It returned HTTP 403 to our client on 2026-08-18, so we did not read it and make no claim about its contents. · retrieved 2026-08-18 - Live handshake against
https://learn.microsoft.com/api/mcp—initializereturned capabilities forlogging,prompts,resourcesandtools, plus a server instructions block describing the three tools and the search/code/fetch workflow;tools/listreturned three tools with full descriptions, input schemas, output schemas and safety annotations;prompts/list,resources/listandresources/templates/listeach returned an empty array (2026-08-18). All requests anonymous. The?maxTokenBudget=2000variant returned the same three tools; the/openai-compatiblevariant returned two, namedsearchandfetch. A plain browser-style GET against the endpoint returned HTTP 405, confirming Microsoft's documented behaviour. Tool schemas committed atdocs/marketing/data/tool-schemas/microsoft-learn.tools.json, byte-identical in content to the live response. · retrieved 2026-08-18 - Auth probe:
https://learn.microsoft.com/.well-known/oauth-protected-resource, the same path suffixed with/api/mcp, and/.well-known/oauth-authorization-serverall returned HTTP 302 redirects into the Learn locale-routing chain rather than an RFC 9728 descriptor (2026-08-18). No OAuth protected-resource document exists, consistent with the successful anonymous handshake and with Microsoft's own no-authentication statement. · retrieved 2026-08-18 - Anthropic Connectors Directory entry — . Tier, categories, endpoint, transport, auth posture and all three tool names, read from our committed directory snapshot dated 2026-08-16. · retrieved 2026-08-16
Use in Agentman
Connect once and your agents call these tools on their own — on a schedule, in a workflow, with nobody at the keyboard.
Open in Agentman StudioServer Info
- Category
- Developer Tools
- Developer
- Microsoft Learn
- Tools
- 3
- Domain
- learn.microsoft.com
Using Claude Desktop or another MCP client? Setup docs — the connection URL above works anywhere.