Boundary 1.1 improves target discovery, external app integration, and introduces MCP server

Boundary 1.1 improves integration with app tokens, simplifies cross-account AWS host discovery, and introduces an MCP server.

Van Phan

Oct 23, 2026

Van Phan

Boundary 1.1 is now available with administrative improvements for enterprise scale and simpler external app integration. A new Model Context Protocol (MCP) server will also be available in public preview that allows administrators and end users to interact with Boundary using natural language through AI-driven interfaces.

»App Tokens

This release introduces app tokens, which are used to securely integrate external tools with scoped access to Boundary. Prior to this release, when an external machine or application needed access to Boundary for CI/CD workflows, infrastructure automation, observability, or any type of integration, it relied on a shared human credential or an admin-managed static password, which posed a security concern as password accounts are long lived and never expire. In addition, when human password credentials were used, the authorization tokens given to external applications were short-lived and valid for only seven days. This means applications would only be able to perform API calls to Boundary for seven days before the application needed to re-authenticate and request a new token. This was an inefficient process that was brittle and not very secure.

App tokens now allow administrators to provision tokens with fine-grained permissions for machine-to-machine communication with Boundary. App tokens will have maximum time-to-live (TTL) ceilings of three years to ensure credentials are not long-lived. In addition, each individual app token will have customizable TTL and time-to-stale (TTS) values so new tokens are not required as frequently. For example, administrators can set tokens to expire after one year instead of every seven days. This capability makes it more efficient for administrators to manage the full credential lifecycle, including creating, revoking, and rotating credentials. Comprehensive audit logs also record every API request, helping organizations meet compliance and regulatory requirements.

»Cross-account support for AWS auto-host discovery

Boundary dynamic host catalog (DHC) automatically discovers, onboards, updates, and off-boards virtual machines (VMs) from Amazon Web Services (AWS), Microsoft Azure, and Google Cloud to keep Boundary’s host inventory current. This automation eliminates manual inventory updates, which are difficult to sustain at scale in environments where VMs are frequently created and removed.

Before this release, DHC could discover EC2 instances only within the AWS account in which it was configured. Discovering instances across multiple accounts required administrators to configuring a worker separately in each AWS account, increasing setup time and ongoing management.

Boundary 1.1 adds cross-account discovery for DHC on AWS. Administrators can configure a DHC in one AWS account and use appropriate AWS Identity and Access Management (IAM) roles and permissions to discover EC2 instances in other accounts. This hub-and-spoke model centralizes configuration and reduces administrative effort.

»MCP server (public preview)

Boundary now offers a Model Context Protocol (MCP) server that simplifies management and troubleshooting on the control plane, while reducing the complexity of integrating AI agents with Boundary for secure infrastructure access on the data plane.

How this helps administrators

As organizations adopt new products or tools, the learning curve associated with managing the product can be a challenge. Even simple products may have multiple tools for administrators to learn, including GUIs, CLIs, and APIs. Further, troubleshooting can be more frustrating as administrators interpret unfamiliar error message. Administrators will often need to search through documentation which can also be a daunting.

The Boundary MCP server now relieves a lot of the burden allowing administrators to interact with Boundary clusters using natural language with common AI assistant clients like Claude, Cursor, and Codex. Administrators can ask questions, perform administrative tasks, and troubleshoot issues without needing to know the exact UI workflow, CLI command, or API call. This reduces the level of effort and time administrators spend learning Boundary tools or searching through documentation.

How this helps AI agents

Integrating agents with an infrastructure access platform can traditionally require developers to build custom integrations using APIs or SDKs, or install and invoke CLI tools within the agent’s environment. The Boundary MCP server simplifies this experience by exposing Boundary’s access capabilities, allowing MCP-compatible agents to request infrastructure access without developers having to build and maintain custom Boundary integrations.

When an agent needs to connect to a resource, the Boundary MCP server can establish the authorized Boundary session and create a local TCP listener that proxies the agent’s connection through Boundary to the target resource. The agent can then connect its existing client or tool to the local listener, while Boundary handles the underlying access session. This means developers do not need to embed Boundary-specific API calls, SDKs, or CLI commands into each agent. Instead, organizations can provide a standardized way for agents to access infrastructure through Boundary while retaining Boundary’s centralized authorization and access controls.

»Conclusion

Boundary 1.1 helps organizations adopt AI agents and other non-human identities without sacrificing control over infrastructure access. JWT authentication and app tokens provide purpose-built identity and credential controls, while the Boundary MCP server simplifies administration and agent integration. Cross-account AWS host discovery further reduces configuration effort at enterprise scale. Together, these capabilities help teams securely expand automation while maintaining centralized authorization, credential management, session control, and auditing.