Plan

The plan server supports plan-first workflows in Aether. It exposes:

• a plan MCP prompt for putting an agent into planning mode
• a submit_plan tool for sending a finished markdown plan to the client for approval or feedback

See Tools for configuration.

Prompt

When the server is enabled it advertises a plan prompt. The prompt accepts a single ARGUMENTS parameter containing the task to plan.

By default the prompt returns Aether’s built-in planner instructions. You can override that behavior with --prompt-file <path> to load your own markdown prompt body instead.

Note

If the configured prompt file is missing at runtime, the server falls back to the built-in default planner prompt.

Tools

Tool	Description
submit_plan	Submit a markdown plan file for review and approval.

submit_plan

Input

Field	Type	Description
planPath	string	Absolute path to the markdown plan file to review.

Output

Field	Type	Description
approved	boolean	true when the reviewer approved the plan.
feedback	string | null	Optional feedback from the reviewer, typically present on denial.

Example request

{
  "planPath": "/absolute/path/to/docs-site-refresh-plan.md"
}

Example responses

{
  "approved": true,
  "feedback": null
}

{
  "approved": false,
  "feedback": "Split the rollout and testing steps into separate sections."
}

Configuration

Enable the built-in plan server in .aether/mcp.json:

.aether/mcp.json

{
  "servers": {
    "plan": {
      "type": "in-memory"
    }
  }
}

To customize the planning prompt:

.aether/mcp.json

{
  "servers": {
    "plan": {
      "type": "in-memory",
      "args": ["--prompt-file", ".aether/prompts/plan-mode.md"]
    }
  }
}

Typical review flow

1. The planner writes a markdown file such as feature-plan.md.
2. The planner calls submit_plan with the absolute file path.
3. The client shows an approve / deny review form, including the plan path and rendered markdown.
4. The tool returns { approved, feedback } so the planner can either continue or revise the plan.

This makes plan approval part of the MCP workflow instead of relying on free-form chat alone.