acknowledge_alert_event
ChatGPTAcknowledge a firing alert event on behalf of the signed-in user. The event stays open and still auto-resolves. Only delegated (user) principals can acknowledge; service accounts receive validation_error. Requires scope: mcp:failures:write.
explain_job_failure
ChatGPTDeterministic failure triage for one job: picks the strongest signal (failedReason, latest stacktrace, or last log lines), lists related alert events, and rates confidence. Stacktraces and logs need mcp:logs:read; related alerts need mcp:failures:read. Missing evidence is reported in skippedSources rather than failing the call. Requires scopes: mcp:diagnostics:read, mcp:jobs:read. Optional scopes that add detail: mcp:logs:read, mcp:failures:read.
find_job
ChatGPTLocate a job by id when the queue is unknown. Scans up to 100 discovered queues on the connection and returns every queue containing that job id. Prefer get_job when the queue is already known. Requires scope: mcp:jobs:read.
get_alert_event
ChatGPTDetail for one alert event including acknowledgement, notification timing, and notification deliveries (channel, status, attempts, last error, external link). Delivery targets are omitted. Requires scope: mcp:failures:read.
get_alert_rule
ChatGPTDetail for one alert rule plus its five most recent events. Requires scope: mcp:failures:read.
get_alert_summary
ChatGPTOpen alert counts for a connection split by acknowledgement, broken down by queue and by rule, plus rule state counts (active, snoozed, disabled). Cheap first call when asked "is anything alerting". Requires scope: mcp:failures:read.
get_connection_overview
ChatGPTOne-call health check for a connection: discovery status, aggregated job counts across up to 100 queues, paused queues, queues with backlog but no workers, top queues by failed and waiting jobs, total workers, and (when scopes allow) open alert counts and the latest Redis health sample. Use it first when asked "how is this system doing". Requires scope: mcp:jobs:read. Optional scopes that add detail: mcp:failures:read, mcp:diagnostics:read.
get_failure_events
ChatGPTAlert events (incidents) for a connection, newest first. Filter by queue, job id, status (firing, resolved, suppressed), acknowledgement, or rule. Suppressed events are informational cooldown records. Requires scope: mcp:failures:read.
get_job
ChatGPTFull detail for one job: state, attempts, payload (data), options, progress, return value, failure reason, and stacktrace count. Sensitive keys and inline secrets are redacted and long strings truncated. Requires scope: mcp:jobs:read.
get_job_logs
ChatGPTPaginated log lines written by the worker via job.log(). Oldest first; use the cursor to page toward the newest lines. Requires scope: mcp:logs:read.
get_job_stacktraces
ChatGPTStacktraces recorded for each failed attempt of a job, newest attempt first. attemptNumber counts from 1; isLatest marks the most recent failure. Requires scope: mcp:logs:read.
get_queue
ChatGPTDetail for one queue: paused state, job counts by state, number of job schedulers, and the workers currently attached (id, address, age, idle time). Requires scope: mcp:jobs:read.
get_queue_metrics
ChatGPTThroughput metrics for one queue over a window of up to 1440 minutes: completed/failed counts, success and failure rates, per-minute averages, longest streaks, lifetime totals, estimated drain time, live counts, and worker/scheduler counts. Requires scope: mcp:diagnostics:read.
get_redis_health
ChatGPTRedis server health for a connection: latest sample (memory %, CPU %, fragmentation, clients, blocked clients, evictions/min, rejected connections/min), active redis_health alert thresholds, and a bounded time series aggregated by max per bucket. Requires scope: mcp:diagnostics:read.
get_scheduled_job
ChatGPTDetail for one job scheduler including its job template (name, data, options), schedule, next run, and recent failure statistics. Requires scope: mcp:jobs:read.
get_workers
ChatGPTWorkers attached to queues on a connection with age and idle time. Pass queueName for one queue, or omit it to page through every queue (up to 50 queues per call). Requires scope: mcp:jobs:read.
list_alert_rules
ChatGPTAlert rules configured for a connection with type, thresholds (config), queue filters, cooldown, state (active, snoozed, disabled), notification channels (secrets removed), and open event counts. Requires scope: mcp:failures:read.
list_connections
ChatGPTList the Redis connections the caller may access, with name, environment, BullMQ key prefix, and organization. Start here to open the interactive queue explorer in MCP Apps hosts, or get a text inventory in other clients. Data tools need a connectionId from this list. Follow nextCursor for all visible connections. Requires scope: mcp:jobs:read.
list_jobs
ChatGPTPaginated job summaries for one queue. Filter by state (status), a case-insensitive substring of the job name (name), or an exact job id (jobId). Name filtering scans up to 10,000 jobs, so combine it with status on large queues. Requires scope: mcp:jobs:read.
list_queues
ChatGPTList discovered BullMQ queues on a connection with paused state and live job counts (waiting, active, delayed, completed, failed, paused, prioritized). Paginated and alphabetical. Requires scope: mcp:jobs:read.
list_scheduled_jobs
ChatGPTBullMQ job schedulers (repeatable jobs) with cron pattern or interval, timezone, next run, iteration count, and recent failure counts. Pass queueName for one queue, or omit it to walk the whole connection (up to 25 queues or pageSize schedulers per page; large queues are split across pages). Requires scope: mcp:jobs:read.
pause_queue
ChatGPTPause a queue so workers stop picking up new jobs. Active jobs finish normally. Safe to call on an already paused queue (changed=false). Requires scope: mcp:queues:pause.
ping
ChatGPTHealth check for the MCP transport. Returns the text "pong". Requires only mcp:discover; use it to confirm authentication and connectivity before calling other tools. Requires scope: mcp:discover.
promote_job
ChatGPTMake a delayed BullMQ job eligible for processing now. Workers, queue pause state and processing limits still determine when it runs. Only jobs in the delayed state can be promoted; other states return conflict. Requires scope: mcp:jobs:promote.
render_ui
ChatGPTOpen an interactive Durabull view for the user. Use ordinary data tools for investigation, intermediate steps and verification; they return text and structured evidence without rendering a UI. Call render_ui only when intentionally presenting an interactive view to the user: when they ask to open/show a UI, or when a chosen view materially helps explain the answer. Prefer one focused view after gathering evidence, not a card for every tool call. Omit render_ui for background work, routine progress and text-only answers. In hosts without MCP Apps, call the selected read tool directly and summarize its evidence. Pass toolName as the read tool for the desired view and arguments exactly as that tool expects; omit both to browse connections. Example: {"toolName":"get_queue","arguments":{"connectionId":"<id>","queueName":"emails"}}. Returns only a validated view request, not fetched evidence or proof that the host displayed it. The UI loads fresh data through the selected read tool with its normal scopes and permissions. Never use this tool to perform a mutation. Requires scope: mcp:discover.
resolve_alert_event
ChatGPTMark a firing alert event as manually resolved. Suppressed events cannot be resolved. Linked external issues (for example Linear) are closed asynchronously. Requires scope: mcp:failures:write.
resume_queue
ChatGPTResume a paused queue. Safe to call on an active queue (changed=false). Requires scope: mcp:queues:pause.
retry_job
ChatGPTRe-enqueue a failed job with its existing payload. Only jobs in the failed state can be retried; other states return conflict. Never modifies job data. Requires scope: mcp:jobs:retry.
snooze_alert_rule
ChatGPTSilence an alert rule for a number of minutes (1 to 10080). Open incidents stay frozen and re-evaluate when the snooze expires. Does not disable the rule. Requires scope: mcp:failures:write.
unacknowledge_alert_event
ChatGPTRemove the acknowledgement from a firing alert event so it shows as unhandled again. Requires scope: mcp:failures:write.
unsnooze_alert_rule
ChatGPTClear a snooze so the alert rule evaluates on the next monitor poll. Requires scope: mcp:failures:write.