Security & Permissions

Architectural context should remain under your control.

Making architecture information available to AI agents introduces an important responsibility: access must remain deliberate, scoped, and revocable.

SPECTRALE's Agent Gateway is designed around controlled access to project information.

Project-scoped access

An authorized agent should only be able to retrieve information from projects explicitly included in its authorization.

Access to one project must not imply access to unrelated projects.

Tool-level permissions

Different workflows require different capabilities.

An agent that only needs component information should not automatically receive access to every available MCP tool.

Tool permissions allow the connection's capabilities to be limited to the intended use case.

Organization boundaries

SPECTRALE organizes access around users, organizations, and projects.

MCP requests must respect the same organizational boundaries as the application.

An agent connection is not an alternative path around normal project authorization.

Credential separation

MCP access credentials and third-party provider credentials serve different purposes.

MCP credentials authorize access to SPECTRALE's MCP resource.

Provider credentials, where applicable, authorize access to an external platform.

These credential types must remain separate.

Read-only tools

The current MCP toolset is focused on retrieving information.

It does not expose general-purpose infrastructure modification capabilities.

Revocation

Agent access should be revocable when a connection is no longer needed.

Revocation is an essential part of maintaining control over automated access.

Operational security

A secure remote MCP deployment also requires controls such as:

• Token expiration and validation.
• Request and execution limits.
• Audit logging.
• Secure secret storage.
• Transport security.
• Input validation.
• Protection against cross-project data access.

SPECTRALE applies these controls to MCP access so architectural context remains scoped, temporary, and auditable. This documentation does not claim independent certifications or audit attestations.

Deployment instructions and authorized agent execution are covered separately in Deployment Security. The read-only context tools on this page are unchanged.