MCP: `get_tasks` returns full task content, overloading AI agent context

[JanJetze]

Problem

The get_tasks MCP tool returns the complete content of every task, including all fields like description, context_summary, working_files, blockers, handoff_notes, activity_summary, etc.

When listing 50+ tasks, this generates a massive response that:

1. Overloads AI agent context - Claude Code shows warnings about context size
2. Wastes tokens - Most of the detailed content is never used when just browsing/filtering tasks
3. Slows down responses - Large payloads take longer to process

Example response for a single task:

{
  "id": "task-7de80f57",
  "story_id": "story-6c734053",
  "title": "Add Recognizer field to Dependencies struct",
  "status": "todo",
  "task_type": "feature",
  "estimate_hours": 0.5,
  "description": "In services/genie/face/internal/deps.go, add `Recognizer *goface.Recognizer` field...",
  "context_summary": null,
  "working_files": null,
  "blockers": null,
  "handoff_notes": null,
  "progress_percent": 0,
  "actual_hours": null,
  "created_at": "2026-02-03T19:27:51.283Z",
  "updated_at": "2026-02-03T19:27:51.283Z",
  "created_by": null,
  "activity_summary": null,
  "story_title": null,
  "epic_id": null,
  "epic_title": null,
  "epic_priority": null
}

Multiply this by 50-100 tasks and the response becomes enormous.

Expected Behavior

get_tasks should return a summary view by default:

{
  "id": "task-7de80f57",
  "title": "Add Recognizer field to Dependencies struct",
  "status": "todo",
  "task_type": "feature",
  "estimate_hours": 0.5,
  "story_id": "story-6c734053",
  "progress_percent": 0
}

The full task details should only be fetched via get_task(task_id) when needed.

Suggested Solutions

Option A: Default to summary, add full parameter

get_tasks(status: "todo", full: true)  // returns all fields
get_tasks(status: "todo")              // returns summary only

Option B: Add fields parameter

get_tasks(status: "todo", fields: ["id", "title", "status", "description"])

Option C: Separate endpoints

• get_tasks - always returns summary
• get_tasks_full - returns complete content (for specific use cases)

Impact

This is particularly important for AI agent workflows where context window management is critical. The current behavior forces agents to consume significant context just to get a task list, leaving less room for actual work.

[srstomp]

@JanJetze

This is already fixed. The fields parameter was implemented in v0.13.0 (option B from the issue). get_tasks now defaults to "minimal" fields, and you can pass fields: "standard" or fields: "full" when needed. The same was also added to get_task.