Loading ToolBench
Exposes Conductor's REST API as MCP tools for Claude. Manages scheduled workers, task chains, and worker groups.
Conductor MCP exposes 18 tools with complete input schemas and descriptions. However, critical gaps limit production readiness: (1) NO output schemas documented, LLMs cannot predict response structure or chain tools; (2) parameter descriptions are sparse or missing for many fields (e.g., 'config' in create_worker_from_template has no guidance); (3) no error handling guidance, tools return raw API errors without recovery hints; (4) destructive operations (delete_all_workers, delete_worker, delete_chain) lack confirmation/dry-run patterns; (5) descriptions are functional but generic, averaging ~120 chars, below the 194-char baseline for A+ tools. Tool naming is strong (verb_noun pattern, clear intent), and schemas are well-formed with enums and required fields. Composition is sound, tools are single-responsibility and chainable. However, missing output documentation and error guidance prevent confident LLM usage.
Create a new task chain. Steps run sequentially by default. Assign same stage number to multiple steps to run them in parallel. Stage numbers are integers 0-99.
Create a new worker group.
Register a new scheduled worker. sched_type: 'fixed' (comma-separated times like '09:00,14:30'), 'interval' (like '2h 30m' or '1h'), or 'cron' (like '0 9 * * 1-5'). timeout_minutes=0 means no timeout. env_vars is KEY=VALUE lines.
Create a worker using a built-in template. Template types: folder_backup (source, dest, keep, summary_email), file_cleanup (folder, pattern, days, summary_email), folder_watcher (watch, rules as [{ext, dest, email_to}]), uptime_check (url, log_file, alert_email), open_url (url), run_and_email (script_path, output_file, email_to). Generated scripts use stdlib only.
Delete ALL workers and chains. Irreversible. Use with caution.
Delete a chain and all its steps by ID.
No output schemas documented. LLMs cannot predict response structure, field names, or types. Prevents tool chaining and forces agents to guess what data is available.
Destructive operations (delete_all_workers, delete_worker, delete_chain, delete_group) lack confirmation or dry-run patterns. Agents can irreversibly delete all data without safeguards.
Error handling is absent. Tools return raw API errors without recovery guidance. LLMs cannot determine if errors are retryable, user-fixable, or fatal.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 69 | 2026-07-28+ | v2 |
Delete a group by ID. Unassigns all members but does not delete them.
Delete a worker by ID. Removes from scheduler and database.
Get the last 10 run history entries for a chain.
Get the last 10 run history entries for a worker. Returns triggered_at, trigger_type, success, duration_ms, error_msg.
List all task chains. Returns id, name, schedule, steps with stage numbers, stop_on_failure, paused state, next trigger, last run status.
List all worker groups. Groups are collapsible collections of workers and chains in the dashboard.
List all workers and chains in Conductor. Returns id, name, task_path, schedule, paused state, next trigger time, remaining runs today, last run status, and group assignment for each.
Pause all workers at once.
Toggle pause/resume for a worker by ID. If paused, resumes it. If running, pauses it.
Fire a chain immediately outside its schedule. Logged as MANUAL trigger.
Fire a worker immediately outside its schedule. Logged as MANUAL trigger type.
Update an existing worker by ID. All fields optional except name and task_path which are required.
Parameter 'config' in create_worker_from_template has no description or schema. LLMs cannot determine what fields to pass for each template type.
Descriptions for destructive operations (delete_all_workers, delete_worker, delete_chain, delete_group) are minimal (<50 chars) and lack severity warnings. LLMs may not recognize irreversibility.