Mcp Tool Quick Reference

MCP Tool Quick Reference (symptom → tool call)

When the user reports a symptom, route to the MCP tool below. Format: symptom phrasing → MCP server → specific tool. Cite the routing in your response.

Cron / scheduled-job symptoms

| Symptom | Server | Tool | |---|---|---| | "Did Friday's cron run?" | silentwatch | list_recent_silent_failures | | "Cron exited 0 but no output" | silentwatch | list_recent_silent_failures (matches EXIT0EMPTYSTDOUT) | | "Cron retry storm" | silentwatch | list_recent_silent_failures (matches RETRYSTORM) | | "Specific cron run by ID" | silentwatch | get_run_by_id |

Cost / billing symptoms

| Symptom | Server | Tool | |---|---|---| | "What's my AI bill" | cost-tracker | cost_overview | | "Which agent is burning tokens" | cost-tracker | cost_overview (per-agent breakdown) | | "Am I about to 429" | cost-tracker | predict_429_in_window | | "When does my rate-limit reset" | cost-tracker | predict_429_in_window (returns window-reset time) | | "Cost spike yesterday" | cost-tracker | find_cost_anomalies | | "Cheaper-model recommendations" | cost-tracker | cheaper_routing_recommendations | | "Anthropic billing surprise / HERMES.md / API rate billing" | cost-tracker | predict_429_in_window (precursor catch) |

Security / supply-chain symptoms

| Symptom | Server | Tool | |---|---|---| | "Is this skill safe to install" | skill-vetter | vet_skill | | "Is this third-party plugin safe" | skill-vetter | vet_skill | | "Is this AGENTS.md / CLAUDE.md / .cursor/rules safe" | skill-vetter v1.1 | vet_agent_config | | "Just got a PR adding to .claude/" | skill-vetter v1.1 | vet_agent_config | | "Cursor CVE / Gemini yolo-mode equivalent" | skill-vetter v1.1 | vet_agent_config |

Bash / shell command symptoms

| Symptom | Server | Tool | |---|---|---| | "Should I run this command" | bash-vet | vet_command_chain | | "Agent emitted a chained command" | bash-vet | vet_command_chain chain_mode=true | | "rm -rf with $VAR concern" | bash-vet | vet_command_chain (matches RMUNSETVAR) | | "apt remove '*' glob fear" | bash-vet | vet_command_chain (matches PACKAGE.APTGLOBWIPEOUT) | | "curl | bash" obfuscation | bash-vet | vet_command_chain (matches EXFIL.BASE64PIPESHELL) |

Upgrade / version symptoms

| Symptom | Server | Tool | |---|---|---| | "Should I upgrade to X" | upgrade-orch | check_known_regressions | | "Did Anthropic silently change something" | upgrade-orch | detect_provider_regression | | "CPU spiked after last upgrade" | upgrade-orch | check_known_regressions (then health-mcp for current state) | | "Reasoning effort feels lower" | upgrade-orch | detect_provider_regression (April-23-Anthropic-style fingerprint) |

Output / claim verification symptoms

| Symptom | Server | Tool | |---|---|---| | "Did the agent actually do X" | output-vetter v1.1 | verify_action_outcome | | "Tests pass but I'm not sure" | output-vetter v1.1 | verify_action_outcome 'tests pass' | | "Agent said it committed but I see no commit" | output-vetter v1.1 | verify_action_outcome 'I committed' | | "Codex sandbox-escalation" | output-vetter v1.1 | verify_action_outcome 'I'm read-only' before_snapshot=... (STATEVIOLATEDCONSTRAINT) | | "Code has silent error masking" | output-vetter | find_swallowed_exceptions | | "Agent response feels ungrounded" | output-vetter | verify_response_grounding |

Health / vitals symptoms

| Symptom | Server | Tool | |---|---|---| | "Is my deployment healthy" | health-mcp | health_overview | | "Gateway up" | health-mcp | health_overview (gateway sub-component) | | "Recent errors" | health-mcp | recent_errors | | "OOM history" | health-mcp | health_overview (resource-pressure section) | | "Disk full" | health-mcp | health_overview (resource-pressure section) |

Audit / end-to-end symptoms

| Symptom | Server | Tool | |---|---|---| | "Audit my deployment" | (use /aufgaard:production-audit skill) | calls all 7 MCPs in parallel | | "End-to-end review" | (use /aufgaard:production-audit skill) | same | | "Is everything OK" | (use /aufgaard:production-audit skill) | same |

Routing principles

1. One symptom, one MCP. If the symptom is precise, route to one MCP. Don't fan out unnecessarily.
2. Multiple symptoms, parallel calls. If the symptom touches multiple categories (e.g. "audit my deployment"), call MCPs in parallel using the production-audit skill.
3. Cite the routing decision. When responding, include "via cost-tracker / health-mcp / etc." in the verdict so the user sees the trail.
4. No-data is honest data. If the chosen MCP returns empty, say so plainly: "cost-tracker has no logged calls yet — install the logging shim and re-run."
5. When uncertain, surface the choices. If the symptom maps to 2+ MCPs and you're not sure which, say "This could be P3.4 (cost-tracker) or P10.1 (silentwatch). Run both? Or describe the symptom in more detail."

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: temurkhan13
• Source: temurkhan13/aufgaard
• License: MIT

Install and usage instructions live in the source repository linked above.