Loading ToolBench
Cross-cloud MCP Server: AI agents can discover resources, correlate logs, and diagnose infrastructure issues across AWS, GCP, Vercel, and Cloudflare.
CloudPulse has 4 well-named tools with clear descriptions (avg 180 chars, within baseline 194). All tools start with action verbs (list_, get_, diagnose_, check_). Input schemas are visible and use Zod with type definitions. However, parameter descriptions are sparse, many lack detail on format, constraints, or dependencies. Output schemas are not documented; responses are JSON-stringified without field-level guidance. Error handling is absent from visible code. No tool annotations (readOnlyHint, destructiveHint). Composition is sound: tools are single-purpose and chainable via platform/region parameters.
Query quota limits and current usage across configured cloud platforms. Highlights resources approaching or exceeding their limits (default warning threshold: 80%). Use this to proactively catch Lambda concurrency limits, Vercel plan caps, and similar issues before they cause outages.
Diagnose connectivity issues between a source service and a target resource. Checks Vercel environment variables, AWS Security Group inbound rules on the required port, and external API reachability. Returns a list of diagnostic results with actionable messages.
Fetch logs from multiple cloud platforms (AWS CloudWatch + Vercel) for a given time window and optional trace ID, then merge them into a single chronological timeline. Use this to correlate errors across frontend and backend services.
Scan all configured cloud platforms (AWS, Vercel, GCP, Cloudflare) and return a unified topology of active services including their endpoints and regions. Run this first to understand the infrastructure landscape.
Output schemas not documented. Tools return JSON.stringify(result) with no field-level guidance. LLMs cannot plan downstream calls or extract specific fields without trial-and-error.
Parameter descriptions lack format/constraint details. 'trace_id' has no format hint; 'warn_threshold' has min/max but no guidance on what happens at 100%. 'target_port' is a number with no range (1-65535?).
No error handling guidance in tool implementations. No recovery hints, retryability classification, or actionable error messages. Agents cannot self-correct on failures.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 60 | 2026-07-28+ | v2 |
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). All tools are read-only but this is not declared in the schema, forcing LLMs to infer safety from descriptions.
Result limits not enforced or documented. 'list_cloud_topology' and 'get_correlated_logs' could return hundreds of items, blowing context window. No pagination or cap mentioned.