Loading ToolBench
Prometheus MCP demonstrates solid definition quality with well-structured tool naming, comprehensive schemas using Zod, and clear input parameter descriptions. All 10 tools follow consistent verb_noun naming patterns (prometheus_list_*, prometheus_query*, etc.). However, tool descriptions are brief (typically 5-15 words) and lack guidance on WHEN to use tools, discovery hints, or recovery patterns. Output schemas are inferred but not explicitly documented in descriptions. Tool composition is clean, each tool has one responsibility. No critical security issues detected (no credentials in parameters). The server uses proper Zod schema definitions with descriptions, but descriptions could be enriched to 50-200 char range per production baselines.
Get Prometheus build information
Get all values for a specific Prometheus label
List all available Prometheus labels
List all available Prometheus metrics
List all Prometheus targets
Get metadata for a specific Prometheus metric
Execute a Prometheus query
Tool descriptions are too brief (5-15 words). Rubric baseline is 194 chars avg, with A+ tools at 50-200 char range. Current descriptions lack WHEN-to-use context and discovery hints (e.g., 'Call this first to see available metrics before querying').
Output schemas are not documented in tool descriptions or visible in the source code. LLMs cannot plan downstream calls or know what fields to expect. Rubric requires: 'Document the output schema. LLMs need to know what fields to expect so they can plan downstream tool calls and extract the right data.'
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | C | 67 | <=2025-11-25 | v2 |
| 2026-03-09 | D | 51 | - | v1 |
Execute a Prometheus range query
Get Prometheus runtime information
Get targets for a specific scrape pool
Discovery tools (list_*, describe_*) lack guidance on what structure they reveal and when to call them. E.g., 'List all available Prometheus metrics' should include: 'Call this first before executing prometheus_query to see available metric names and understand the namespace structure.'
No pagination or result-limiting mechanism visible for tools returning lists (prometheus_list_metrics, prometheus_list_labels, prometheus_list_targets, prometheus_scrape_pool_targets). Rubric states: 'Even if the API allows returning thousands of items, cap results at a reasonable limit (e.g. 20-50) and offer pagination.' Large result sets can exhaust context window.
Query tools (prometheus_query, prometheus_query_range) accept RFC3339 or unix timestamps but descriptions do not specify format constraints or examples. Rubric: 'When a parameter has a regex pattern, length limit, or character restriction, state it in the description.' Should clarify: 'start timestamp (RFC3339 format, e.g. 2025-01-15T10:30:00Z, or unix timestamp in seconds)'.