Static source inference · medium confidence · detected: Logging
Deprecated protocol patterns detected
Summary
The SigNoz MCP server has 9 tools with consistent naming patterns (all verb_noun style: fetch_*, test_*, execute_*) and present descriptions. However, the implementation has critical gaps: input schemas are minimally documented with no visible type constraints, parameter descriptions lack depth and actionable guidance, and output schemas are completely undocumented. Tool descriptions are generic and do not follow LLM-optimized patterns. No error handling guidance is visible, and several tools accept free-form parameters (sql, queryType, filters, options) without validation constraints. The server lacks the depth expected for production agent tooling.
No output schemas documented for any tool. LLMs cannot predict what fields to expect or plan downstream tool calls. Tools like signoz_fetch_dashboards should document: returns array of {id, name, createdAt, ...}, total count, pagination cursor.
signoz_execute_clickhouse_query and signoz_execute_builder_query accept free-form SQL and query parameters without input validation or constraints. No enum values, no regex patterns, no length limits. LLMs will hallucinate invalid queries.
Add output schemas to all tools. Document the response structure as: 'Returns array of objects with fields: id (string), name (string), createdAt (RFC3339), labels (object), status (string). Total count: integer. Includes next_cursor for pagination if >50 results.'
Replace free-form parameter descriptions with explicit constraints. For signoz_execute_builder_query aggregation param: 'One of: avg, sum, min, max, count, rate, histogram_quantile'
For signoz_execute_builder_query filters: 'Array of filter objects. Each: {field: string, operator: one of (eq, neq, gt, gte, lt, lte, contains, regex), value: string|number}. Example: [{field: "status", operator: "eq", value: "active"}]'
Add step parameter format guidance to signoz_execute_builder_query: 'Time interval for aggregation. Format: number + unit. Allowed units: s (second), m (minute), h (hour), d (day). Examples: 1m, 5m, 1h. Min: 1s, Max: 30d.'
Document relationship between dashboardID and dashboardName: 'signoz_fetch_dashboard_details requires dashboardID. Call signoz_fetch_dashboards first to discover IDs by name, or use signoz_fetch_dashboard_data if you know the name.'
Add error recovery guidance: 'If dashboard not found, returns 404 with available dashboard names. Call signoz_fetch_dashboards to list all and retry with correct ID.'
Add pagination to signoz_fetch_dashboards, signoz_fetch_services, signoz_fetch_traces_or_logs: include limit (default 20, max 100) and cursor/offset parameters. Return total count and next_cursor in response.
Spec posture evidence
Inferred effective spec: <=2025-11-25.
Relies on Logging (deprecated) - log to stderr or use OpenTelemetry
Score history
Overall score trend
↑ 14 points across a rubric change (v1 → v2)
45/100
Scored
Grade
Overall
Spec posture
Rubric
2026-09-22
F
45
<=2025-11-25
v2
2026-03-09
F
31
-
v1
read onlyauthsource verified60/100
Fetch a list of all services monitored by SigNoz with their metrics and performance data
Test the connection to a SigNoz instance and verify it is accessible
Parameter descriptions are generic and lack actionable guidance. E.g. 'The name of the dashboard to fetch data for' does not explain: Is it the display name or internal ID? Can it be partial? What if multiple dashboards match? Should I call signoz_fetch_dashboards first?
signoz_fetch_services has optional startTime and endTime parameters with no description of their purpose (metric aggregation window? service discovery window?). No format constraints, no guidance on defaults.
signoz_execute_builder_query has complex input (queryType, metricName, aggregation, filters array, groupBy array, step, options object) but no examples, no constraints on valid aggregation functions (avg, sum, max, min?), no documentation of filters array format, no guidance on step interval syntax.
No error handling guidance documented. If a dashboard ID is invalid, what error is returned? Can the LLM retry? Should it list dashboards and try again? No recovery guidance visible.
signoz_fetch_dashboard_data accepts dashboardName but signoz_fetch_dashboard_details accepts dashboardID. No guidance on how these relate or which to use when. Forces LLM to guess or make extra lookup calls.
No pagination parameters documented. If signoz_fetch_services or signoz_fetch_dashboards return 1000+ items, how does the LLM limit results? No limit, offset, page_size, or cursor parameters visible.
Document time range constraints for all time-based tools: 'startTime and endTime must be RFC3339 formatted (e.g. 2024-01-15T10:30:00Z). startTime must be before endTime. Max range: 90 days. If omitted, defaults to last 24 hours.'
Add tool descriptions that explain WHEN to use each tool: 'Use signoz_fetch_apm_metrics to get latency, error rate, request throughput for a service. Use signoz_fetch_dashboard_data to render dashboard panels with time-series data. Use signoz_execute_builder_query for custom metric queries not captured in dashboards.'
Document security constraints for signoz_execute_clickhouse_query: 'Only SELECT queries allowed. INSERT, UPDATE, DELETE, ALTER rejected. Timeout: 30 seconds per query. Max result rows: 10,000.'
Add parameter constraints to signoz_fetch_traces_or_logs limit: 'Integer, min: 1, max: 1000, default: 50. Higher limits may timeout; use pagination cursor for large result sets.'
For signoz_test_connection, add: 'Returns {status: "ok", version: string, responseTime_ms: number}. Call this before other tools to verify connectivity. If returns error, check SIGNOZ_MCP_SIGNOZ_URL environment variable.'