Orchestrates CLI agents (Claude Code, Kiro, Codex, Cursor) as task executors in isolated worktrees. Provides MCP tools for managing agent runs, tasks, worktrees, and metrics.
Server has 23 tools with basic schemas and descriptions, but significant gaps limit production readiness. Most tools have descriptions (10-150 chars, below the 194-char baseline), and input schemas are present but sparse. Parameter descriptions are minimal or absent for many tools. Error handling is present but lacks recovery guidance. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk levels. Output schemas are not documented. The server exposes complex orchestration logic (worktrees, roles, isolation modes) but does not guide LLMs on when to use each variant.
Cancel a running run
Commit changes from a run's worktree to git
Create a new orchestration task
Delete a worktree by path
Delete the worktree associated with a run
Continue conversation with an agent in the same worktree
Collect and optionally delete old or dirty worktrees
Get current budget and queue status
Get server health status including uptime, PID, queue and budget status
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk levels. Tools like discard_run_worktree, delete_worktree, and garbage_collect_worktrees are marked DESTRUCTIVE but lack schema hints to guide LLM caution.
Output schemas not documented. Tools return complex objects (runs with events, diffs, worktree state) but LLMs have no schema to understand what fields to expect or how to chain calls.
start_run has 14 optional parameters with minimal descriptions. LLMs cannot determine which combinations are valid (e.g., when to use 'role' vs 'mode', or when 'isolation' applies). Parameter relationships undocumented.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 51 | 2026-07-28+ | v2 |
Get usage metrics for a time period
Get information about a git repository including branch and worktrees
Get details of a specific run by ID
Get file diff for a run's worktree changes
Get event log for a run, optionally from a specific sequence number
Get details of a specific task by ID
List available browser presets for QA testing
List available LLM models for a tool or all tools
List all runs, optionally filtered by task ID
List all orchestration tasks
List available agent tools (Claude Code, Kiro, Codex, Cursor) with their capabilities, session counts, and live sessions
List all worktrees with their state and associated run information
Execute a task (orchestration flow)
Start a new agent run with a prompt, tool, and optional isolation/mode/model parameters
Error handling lacks recovery guidance. Tools return generic error messages without suggesting next steps (e.g., 'unknown tool' does not list available tools; 'run not found' does not suggest list_runs).
Enum constraints missing for string parameters. 'tool' param in start_run accepts 'claude|kiro|codex|cursor' but is not declared as an enum; 'mode' accepts 'ro|verify|rw|full' but lacks enum constraint. LLMs may hallucinate invalid values.