Anthropic Launch Enterprise-Managed Auth for MCP Connectors

Anthropic has launched Enterprise-Managed Authorization for Model Context Protocol (MCP) connectors, giving organizations a centralized way to control how employees connect Claude to workplace applications. The feature allows administrators to provision supported MCP connectors through their existing identity provider, beginning with Okta, so employees can receive approved connections automatically when they sign in rather than completing separate authorization steps for every application. Anthropic initially introduced the feature in beta in June and said on August 24 that enterprise-managed authorization is now generally available.

The update addresses one of the biggest obstacles to deploying AI agents across large organizations: managing authentication, permissions and revocation when AI systems need access to business tools. Under the new system, administrators can authorize connectors once, assign access through existing identity groups and roles, and maintain control through the organization’s identity provider. Anthropic says the feature now supports Datadog, Notion and Slack, in addition to earlier integrations including Asana, Atlassian, Canva, Figma, Granola, Linear and Supabase, with Exa, Miro and Zoom coming soon.

Anthropic Brings Enterprise-Managed Authentication To MCP

The Enterprise-Managed Authorization extension is designed to make MCP connectors easier to deploy at organizational scale.

MCP is an open protocol originally introduced by Anthropic to allow AI assistants to connect with external data sources, business applications and development tools. Instead of creating a separate integration architecture for every AI application, MCP provides a standardized way for AI systems to interact with external tools and information.

Anthropic’s latest authorization layer adds centralized identity management to those connections.

Enterprise-Managed Auth At A Glance

Feature	Details
Company	Anthropic
Technology	Model Context Protocol
New capability	Enterprise-Managed Authorization
Availability	Generally available as of Aug. 24, 2026
Initial identity provider	Okta
Primary users	Enterprise IT administrators
User experience	Automatic connector access on first login
Access management	Identity-provider groups and roles
Supported Claude surfaces	Claude, Claude Code and Cowork
Key benefit	Centralized connector provisioning and revocation

The change effectively moves MCP authorization from an individual-user process toward an enterprise identity-management process.

How Enterprise-Managed Auth Works

Previously, an organization could enable an MCP connector for employees, but individual users still had to authorize that connector themselves.

That created an additional step for every employee and made it harder for IT departments to maintain a consistent access policy.

Under enterprise-managed authorization, the organization connects its identity provider to Claude and determines which connectors employees or groups can use.

OLD MODEL

IT Enables Connector
        ↓
Employee Opens Claude
        ↓
Employee Authorizes Connector
        ↓
Access Granted


NEW MODEL

IT Connects Identity Provider
        ↓
Admin Authorizes Connector
        ↓
Assigns Groups / Roles
        ↓
Employee Logs In
        ↓
Connector Already Available

Anthropic describes this as zero-touch connector setup for end users. Employees inherit access through the identity groups and roles they already have.

Identity Providers Become The Control Point

The major architectural change is that MCP access can now be governed through the same identity infrastructure companies already use for other applications.

Okta is supported at launch, while Anthropic says additional identity providers are coming.

This means IT teams do not need to create a separate authorization-management process specifically for Claude connectors.

Enterprise Access Flow

Step	Administrator Action	Employee Experience
1	Connect identity provider	No action
2	Select MCP connectors	No action
3	Assign groups or roles	No action
4	Employee signs into Claude	Connectors appear automatically
5	Employee uses approved tools	No repeated authorization
6	Employee leaves organization	Access can be revoked through IdP

This approach can be particularly important for companies with thousands of employees and dozens of enterprise applications.

Faster Revocation When Employees Leave

Security teams face a different problem when employees leave an organization.

A conventional access token can potentially remain active after a user’s employment ends unless it is revoked or expires. Anthropic says enterprise-managed authorization allows administrators to use shorter access-token lifetimes because identity-provider checks can occur without adding significant friction for employees.

That creates a more direct connection between an organization’s identity system and AI-tool access.

Employee Deprovisioned
        ↓
Identity Provider Removes Access
        ↓
Claude Connector Access Expires
        ↓
Former Employee
Cannot Continue Using
Enterprise Connector

For companies deploying AI agents against sensitive systems, this type of lifecycle control can become as important as the initial authentication process.

Administrators Can Separate Work And Personal Accounts

Another feature is the ability for administrators to require that a connector be accessed only through the organization’s identity provider.

This is designed to prevent employees from accidentally connecting a personal account to a company-approved workplace connector.

For example, an employee could otherwise have access to both a personal and corporate account for the same service. Centralized authorization gives IT teams a mechanism to ensure that the Claude connection uses the intended enterprise identity.

Access-Control Benefits

Security Requirement	Enterprise-Managed Auth
Centralized provisioning	Yes
Group-based access	Yes
Role-based access	Yes
Centralized revocation	Yes
Shorter token lifetimes	Supported
Work/personal separation	Supported
Individual OAuth setup	Reduced/eliminated for managed connectors

The broader goal is to make MCP connectors behave more like other enterprise applications already governed through centralized identity systems.

MCP Becomes More Important To Enterprise AI

Anthropic’s move comes as MCP expands beyond a developer-oriented integration protocol toward infrastructure for enterprise AI agents.

Anthropic said in July that MCP had surpassed 400 million monthly SDK downloads, representing a fourfold increase during 2026. The company also described MCP as an industry standard for connecting AI agents to applications.

MCP Growth

MCP SDK Downloads

2026
        ↓
400M+ Monthly Downloads
        ↓
4× Growth During 2026
        ↓
Increasing Enterprise Adoption
        ↓
Greater Need For Centralized Security

The more widely MCP is deployed, the more important authorization becomes.

A protocol that allows AI agents to access enterprise systems must provide organizations with a way to determine who can connect, which tools they can use and when that access should end.

Enterprise-Managed Authorization Is An Open MCP Extension

Anthropic is not introducing the authorization mechanism as a proprietary connector system.

Enterprise-managed authorization is an extension to the open MCP authorization specification. The MCP community describes the extension as stable and says organizations can centrally provision MCP server access through their identity providers.

That means other identity providers and MCP providers can potentially implement the same mechanism.

MCP Authorization Ecosystem

Layer	Role
Identity provider	Determines employee identity and permissions
MCP authorization extension	Standardizes enterprise-managed access
MCP connector/server	Provides access to an application
Claude	Uses the approved connection
Enterprise administrator	Defines policies and group access
Employee	Uses tools according to assigned permissions

This standardized approach could reduce fragmentation as the number of MCP-compatible applications grows.

More Enterprise Connectors Are Coming

Anthropic’s connector ecosystem is expanding.

At the original launch, enterprise-managed authorization supported Asana, Atlassian, Canva, Figma, Granola, Linear and Supabase. The August update added Datadog, Notion and Slack, while Anthropic said Exa, Miro and Zoom were coming soon.

Supported And Upcoming Connectors

Status	MCP Connectors
Earlier support	Asana, Atlassian, Canva, Figma
Earlier support	Granola, Linear, Supabase
Newly supported	Datadog, Notion, Slack
Coming soon	Exa, Miro, Zoom

The growing list illustrates why centralized authorization becomes increasingly valuable. Managing three connectors manually is one problem; managing dozens across thousands of employees is another.

Claude Code And Cowork Also Get Consistent Access

Enterprise-managed connector access is not limited to Claude’s conventional chat interface.

Anthropic says access remains consistent across Claude chat, Claude Code and Cowork.

That is important because enterprises increasingly use multiple AI surfaces for different types of work.

Claude may be used for knowledge work, Claude Code for software development and Cowork for other workplace tasks. If all three can use centrally managed MCP connectors, companies can apply a common identity and access policy across those environments.

                 ENTERPRISE IDENTITY
                         │
             ┌───────────┼───────────┐
             ↓           ↓           ↓
          Claude     Claude Code   Cowork
             │           │           │
             └───────────┼───────────┘
                         ↓
                  MCP Connectors
                         ↓
              Enterprise Applications

This creates a more unified architecture for deploying AI throughout an organization.

Security Becomes Critical As AI Agents Gain More Access

MCP connectors can give AI systems access to information and tools that previously required employees to interact with applications directly.

That creates significant productivity opportunities, but it also expands the security surface.

Anthropic has previously emphasized that AI agents need privacy protections and controls when interacting with external systems. Its framework for safe agents specifically notes that MCP connectors include controls allowing users and administrators to determine which tools Claude can access.

The new authorization system adds another layer of enterprise governance.

AI Agent Security Layers

Employee Identity
       ↓
Identity Provider
       ↓
Group / Role Permissions
       ↓
MCP Authorization
       ↓
Connector
       ↓
Enterprise Application
       ↓
Data / Action

The objective is to ensure that an AI agent receives no more access than the employee or organizational role is supposed to have.

Enterprise AI Moves From Experiment To Infrastructure

The introduction of enterprise-managed authorization reflects a broader shift in how businesses are approaching AI.

Early enterprise AI deployments often involved individual employees experimenting with chatbots. As organizations move toward AI agents that can access company data and take actions in business applications, IT departments need the same controls they apply to conventional enterprise software.

Authentication, authorization, identity management, auditability and revocation therefore become foundational components.

From AI Experiment To Enterprise System

Early AI Adoption	Enterprise AI Adoption
Individual users	Organization-wide deployment
Standalone chatbot	Connected AI agent
Manual access	Centralized provisioning
Simple prompts	Tool and data access
Limited permissions	Role-based controls
User-managed accounts	Identity-provider governance

MCP’s evolution is increasingly addressing these requirements.

MCP 2026 Specification Adds More Enterprise Infrastructure

Anthropic’s July MCP specification update also introduced changes designed to make the protocol more suitable for production environments.

The MCP 2026-07-28 release moved the protocol toward a stateless core, standardized extensions and strengthened authorization around OAuth 2.0 and OpenID Connect deployments.

The specification also introduced or formalized MCP Apps and Tasks and added other infrastructure improvements.

Recent MCP Developments

Development	Enterprise Significance
Stateless core	Easier scaling
OAuth/OIDC-aligned authorization	Better identity integration
Enterprise-managed auth	Centralized access
MCP Apps	Interactive application experiences
Tasks	Long-running agent work
Observability	Connector performance monitoring
MCP tunnels	Private-network connectivity

Together, these changes suggest MCP is evolving from a simple connection standard into a broader infrastructure layer for AI agents.

Why Zero-Touch Access Matters For Large Companies

Consider an organization with 10,000 employees and 20 approved AI connectors.

Under a user-by-user authorization model, thousands of employees may have to complete authorization flows individually. IT teams must also deal with employees changing roles, leaving the company or gaining access to new applications.

Centralized provisioning shifts much of that work to the organization’s existing identity infrastructure.

10,000 Employees
       +
20 Enterprise Connectors
       ↓
Potentially Thousands Of
Manual Authorization Events
       ↓
Enterprise-Managed Auth
       ↓
Centralized Provisioning
       ↓
Group-Based Access
       ↓
Automated Revocation

The actual operational savings will vary by organization, but the architectural advantage is clear: AI connector access can follow existing enterprise identity workflows rather than becoming a separate IT process.

The Bigger Picture

Anthropic’s enterprise-managed authorization launch is an important step in the maturation of MCP and enterprise AI agents. As Claude gains the ability to interact with more workplace applications, companies need a way to control those connections without forcing every employee through separate authorization processes. Anthropic’s approach places the identity provider at the center of that process, allowing administrators to provision connectors, assign access through groups and revoke permissions centrally.

The development also shows why authentication is becoming a strategic part of the agentic AI stack. MCP has grown rapidly, and its latest specification is adding production-oriented capabilities around authorization, scaling, applications, long-running tasks and observability. If AI agents are eventually expected to operate across dozens of enterprise applications, standardized identity and permission controls will be essential for making that deployment manageable and secure.

Looking Ahead

Anthropic’s immediate focus will be expanding enterprise-managed authorization beyond Okta and increasing the number of MCP connectors that support the standard. Datadog, Notion and Slack are now supported, while Exa, Miro and Zoom are listed as coming soon. The open nature of the MCP extension also allows other identity and MCP providers to implement the same authorization mechanism.

For enterprises, the bigger question will be how quickly MCP becomes a standard access layer for AI agents across business applications. If adoption continues, centralized identity management could become a basic requirement rather than an optional enterprise feature. Anthropic’s move therefore positions MCP not only as a way for Claude to connect to tools, but increasingly as part of the security and governance infrastructure required to deploy AI agents at organizational scale

Get the day’s top stories in your inbox

One concise email. No spam, unsubscribe anytime.