Claude Swarm
未认领MCP server for orchestrating parallel Claude Code worker swarms with protocol-based behavioral governance, persistent state, and real-time monitoring dashboard
安装
claude mcp add claude-swarm --scope user -- node $(pwd)/dist/index.js工具（40 个）
add_feature
Add discovered work
auto_orchestrate
Hands-free orchestration until completion
check_all_workers
Check all active workers at once
check_reviews
Monitor review worker status
check_rollback_conflicts
Check for conflicts with other workers
check_worker
Get worker output (supports heartbeat mode)
commit_progress
Create git checkpoint
configure_plan_mode
Enable/configure worker plan-before-implement mode
configure_ralph_loop
Enable/configure Ralph Loop execution defaults
configure_reviews
Set auto-review preferences
configure_verification
Set pre-completion verification commands
evaluate_plans
Compare plans and select winner
get_feature_complexity
Analyze complexity score
get_progress_log
View history (paginated)
get_review_results
Get aggregated findings (summary/detailed/json)
get_session_stats
Success rates and timing
get_worker_confidence
Get detailed confidence breakdown
implement_review_suggestions
Convert review findings into features
mark_complete
Mark feature done/failed (auto-retry)
orchestrator_init
Start session with task and features
orchestrator_reset
Clear state and kill all workers
orchestrator_status
Get current state (use after compaction)
pause_session
Pause and stop all workers
pytest
, python -m pytest
resume_session
Resume paused session (supports crash recovery)
retry_feature
Reset failed feature for retry
rollback_feature
Restore files changed by a worker
run_review
Manually trigger code/architecture reviews
run_verification
Run tests/build commands
send_worker_message
Send instructions to running worker
set_confidence_threshold
Configure alert threshold
set_dependencies
Define feature dependencies
setup_analyze
Analyze repo freshness and missing configs
setup_init
Initialize repo configuration with parallel workers
shutdown_session
Graceful shutdown with checkpoint
start_competitive_planning
Spawn 2 planners with different approaches
start_parallel_workers
Launch multiple workers simultaneously
start_ralph_loop
Launch worker with Ralph Loop (fresh context per iteration)
start_worker
Launch worker for a feature
validate_workers
Pre-flight validation before parallel execution