A read-only MCP server that collects multi-source evidence timelines for homelab services from adapters like Gatus, Docker, Loki, Healthchecks, Beszel, and ntfy
Strong foundation with 7 well-defined tools, all with descriptions and input schemas. Tool names follow verb_noun pattern (list_services, search_logs, get_evidence). Descriptions are clear and specific (avg ~120 chars), explaining WHAT each tool does and key constraints. All parameters have type definitions and descriptions. However, output schemas are not explicitly documented in the source code, only inferred from implementation. Error handling is present but minimal (toolError function exists but lacks recovery guidance). No tool annotations visible despite errorReporting=true claim. STDIO transport caps protocol readiness.
Give the server version, active adapters, services with adapters, limits, and compatibility notes. The response does not include base URLs or secrets.
Give Healthchecks checks with a current state of down, grace, or paused. The tool does not include ping URLs.
Give one cached evidence item by its opaque process-local id. The cache has a short TTL.
Collect a multi-source evidence timeline for a service_id with the limit from the configuration. The correlation does not identify a root cause.
Give services from the registry with per-source coverage. Use the optional prefix filter and pagination.
Examine Loki logs for a service_id with the selector from the configuration. Use the optional text filter. The caller cannot change the URL or stream selector labels.
Output schemas not documented in source code. Tool descriptions explain what is returned (e.g., 'multi-source status snapshot', 'evidence timeline'), but structured output format, field names, and types are not visible in the provided code. LLMs cannot plan downstream tool calls without knowing response structure.
Error handling lacks recovery guidance. toolError() function returns errResult(err) but source does not show what errResult produces or whether it includes actionable next steps (e.g., 'Try search_services() first'). Agents cannot self-correct from errors.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 62 | 2026-07-28+ | v2 |
Get a multi-source status snapshot for one service_id (Gatus, Docker, Healthchecks, Beszel). The tool does not get data from Loki or ntfy.
Tool annotations (readOnlyHint, destructiveHint, idempotentHint) claimed in features but not visible in tool registration code. All 7 tools are READ_ONLY per the spec, but this is not declared in the MCP tool definition itself, only in the evaluation metadata.
Parameter constraints not fully specified. search_logs accepts 'regex' with 'built-in limits' but the limits are not documented in the description. incident_context accepts 'max_items' but does not state the server-side maximum or what happens if the value exceeds it.
Pagination support incomplete. list_services accepts offset/limit but does not document total count or next_cursor in the response. Without knowing the total, agents cannot determine if more results exist or plan multi-page retrieval.