Deploy on Agent

Deploy on Agent connects SPECTRALE's visual architecture editor to an external AI agent through the Model Context Protocol (MCP).

The feature separates architectural intent from execution: SPECTRALE describes the requested changes, while the connected agent determines how to implement them using its available tools.

Before you deploy

Make sure that:

1. Your project has a prepared architecture draft.
2. The draft contains at least one effective semantic architecture change.
3. An AI agent is connected and authorized for the relevant project and deployment capabilities.
4. Your Architect account has sufficient deployment tokens.
5. The selected agent has access to the systems and tools required for the intended work.

Step 1 — Review your changes

Open Deploy on Agent from the workspace.

Review the proposed architecture modifications before submission.

The deployment request describes the intended changes, not a guarantee that every operation can be performed automatically.

Step 2 — Select an authorized agent

Choose an agent from the list of compatible, authorized agents.

An agent must be connected and have the necessary deployment permissions before it can receive the request.

Agent authentication and deployment authorization are separate concerns.

A successfully authenticated connection does not automatically authorize every deployment.

Step 3 — Create the deployment request

When you confirm the deployment, SPECTRALE prepares a structured, machine-readable deployment instruction set.

This instruction set is referred to as an Architecture Change Manifest (ACM).

The manifest represents the intended architecture transition and provides context for the selected agent.

Architecture Change Manifest

An ACM can describe information such as:

• The project and deployment context
• The intended architecture changes
• Components and relationships affected by the changes
• Relevant identifiers and dependencies
• Execution constraints and preconditions
• Verification expectations

The exact manifest fields and supported operations depend on the deployed ACM schema version.

The manifest is designed to communicate architectural intent consistently to compatible agents.

It is not an executable program and should not be treated as unrestricted permission to change infrastructure.

Step 4 — Agent retrieves the task

SPECTRALE makes the deployment request available through its MCP integration.

The authorized agent retrieves the task using the supported MCP workflow.

The agent then interprets the requested changes and uses its own tools to perform the work.

SPECTRALE does not remotely launch the external AI agent.

Step 5 — Monitor execution

The agent can report execution progress and results through the supported deployment communication tools.

Users can review updates in the deployment conversation.

If the agent needs clarification, the workflow can support user-agent communication before work continues.

Progress in the deployment view is refreshed on an interval. It is not a continuous event stream.

Deployment tokens

Under the current token model:

One accepted deployment request consumes one deployment token.

Preparing drafts and making layout-only changes do not consume deployment tokens.

Architect currently includes 25 deployment tokens each billing period. Additional packages are 25 tokens for €15. Other plan allocations are pending approval and are not offered as final amounts.

Refer to the current Pricing page and account billing interface for the applicable allowance and package details.

Important limitations

The selected agent must have the necessary external permissions to perform the work.

A deployment request may fail or require user intervention.

Infrastructure changes may have operational consequences and should be reviewed before execution.

Continue with Deployment Security to understand the authorization and instruction-protection model.