Live
Simplify Model Management with Cloudflare AI Gateway Model AliasesRunning OpenShell Agents on Kubernetes: Architecture and Ops Guide for Platform TeamsBorderless Lakehouse Updates: Direct Gemini Queries, Cross‑Cloud Caching, and AlloyDB Iceberg AccessGuided Diagnostics in GoAnywhere MFT promises up to 160 hours of admin time saved per yearImplementing Four‑Layer Context Engineering for Production AI Agents with Deep InsightMachine‑Speed Code Security with CodeMender: Practical Implications for AI‑Driven DevelopmentClaude‑issued identity assertions eliminate per‑plugin sign‑in for enterprise usersAgentic Helpdesk Architecture on Amazon Bedrock AgentCore: Practical Design and Ops TakeawaysSimplify Model Management with Cloudflare AI Gateway Model AliasesRunning OpenShell Agents on Kubernetes: Architecture and Ops Guide for Platform TeamsBorderless Lakehouse Updates: Direct Gemini Queries, Cross‑Cloud Caching, and AlloyDB Iceberg AccessGuided Diagnostics in GoAnywhere MFT promises up to 160 hours of admin time saved per yearImplementing Four‑Layer Context Engineering for Production AI Agents with Deep InsightMachine‑Speed Code Security with CodeMender: Practical Implications for AI‑Driven DevelopmentClaude‑issued identity assertions eliminate per‑plugin sign‑in for enterprise usersAgentic Helpdesk Architecture on Amazon Bedrock AgentCore: Practical Design and Ops Takeaways
Anthropic

Claude‑issued identity assertions eliminate per‑plugin sign‑in for enterprise users

AI SummaryPowered by AI

Anthropic now lets Claude generate signed identity assertions so enterprise plugins can be auto‑connected without each user manually authorizing them. This shifts part of the authentication flow to Claude, reducing friction for engineers while adding a new trust and operational responsibility.

Anthropic has added a beta feature that allows Claude to issue signed identity assertions on behalf of employees. When an authorized user opens a supported plugin, Claude presents the assertion to the service, which validates it and returns an access token. This removes the need for users to sign in to each plugin individually and lets administrators enable auto‑connect for selected services and roles.

How the new Claude‑issued assertions work

The capability builds on the Enterprise‑Managed Authorization (EMA) extension to the Model Context Protocol (MCP). Previously, EMA required the organization’s identity provider to understand the extension; Okta was the only provider with support at launch. The new flow lets Claude act as the issuer of the signed assertions, so the identity provider does not need EMA support.

Admins configure the connection by approving Claude for the organization, selecting the Claude‑issued option in the connector settings, and registering Claude’s unique issuer URL with the target service. Role‑based enablement is possible – for example, all employees may receive a productivity tool, while only a privileged role receives access to a database‑related plugin. The initial rollout includes more than a dozen services such as Amplitude, Atlassian, Datadog, Dropbox, Figma, HubSpot, Linear, Notion, PagerDuty, and Supabase.

Architectural and implementation considerations

Integrating Claude‑issued assertions requires the following steps:

  • Ensure each employee already has a matching SSO account in the target service.
  • Register Claude’s issuer URL on the service’s MCP server.
  • Map organization roles to the services that should be auto‑connected.
  • Test the connection before enabling production access.

Because Claude supplies the signed assertion, the service still performs its own token issuance after verification. The process handles token renewal automatically, so users do not need to re‑authorize when tokens expire.

Operational and security implications

Claude now functions as a trusted issuer, which introduces a new trust boundary. Operators must protect Claude’s signing keys and monitor the issuance process for anomalies. Revocation is not automatic; removing a user from the corporate identity provider does not invalidate a Claude‑issued assertion. Administrators must delete the user from both Claude and the downstream service to fully revoke access.

The limited adoption of EMA among identity providers means many organizations will rely on Claude as the issuer for the foreseeable future. This reliance should be reflected in audit and compliance reviews, with particular attention to key rotation, assertion validation logic in the service, and the lifecycle of role assignments.

Related CloudNinjas coverage: AI engineering.

What This Means For Practitioners

  • Plan to update plugin connector configurations to use Claude‑issued assertions where auto‑connect is desired.
  • Validate that employee SSO accounts exist in each target service before enabling the feature.
  • Incorporate dual‑deletion procedures into off‑boarding workflows to ensure access is removed from both Claude and the service.
  • Monitor Claude’s signing key management and consider periodic key rotation as part of your security posture.
  • Track broader EMA adoption; if your identity provider later supports EMA, you may choose to shift back to provider‑issued assertions.
Originally published atThe New Stack