This server has serious definition quality gaps. While all 7 tools are explicitly registered with Zod schemas and descriptions, most descriptions are terse (10-40 chars), parameter documentation is minimal, and critical operational guidance is missing. No output schemas are documented. Error handling returns raw JSON without recovery guidance. The tools cover a focused domain (Vercel deployments) but lack the LLM-optimized clarity needed for reliable agent selection and chaining. Naming is acceptable (verb-first, action-clear) but parameter descriptions often omit constraints, formats, and dependencies. None of the tools declare their risk level (read-only vs reversible vs destructive) in the description itself, forcing LLMs to infer intent.
Cancels a deployment
Deletes a deployment
Gets a deployment by ID or URL
Gets deployment events by deployment ID and build ID
Gets deployment file contents
Lists deployments
Lists deployment files
Descriptions too short and lack operational context. Most tool descriptions (10-40 chars) fall below the 50-200 char production baseline. They state WHAT the tool does but not WHEN to use it, what it returns, or prerequisites. Example: 'Gets deployments' omits filtering options, pagination behavior, and common use patterns.
Parameter descriptions are minimal or generic. Many parameters lack constraints, formats, or examples. Example: 'limit' and 'since'/'until' parameters have no stated valid ranges or formats. 'direction' enum is documented but others are not. LLMs cannot infer whether limit=1000 is valid or if 'since' expects unix timestamp vs ISO8601.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 45 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 44 | - | v1 |
No output schemas documented. Tool handlers return raw JSON responses via `JSON.stringify(result, null, 2)` without explaining the response structure. LLMs do not know what fields to expect (e.g., does getDeployment return 'id' or 'deploymentId'? Does it include team context?). Prevents chaining, if getDeploymentFiles should accept a fileId returned by listDeploymentFiles, that dependency is invisible.
Risk classification not visible in descriptions. Three tools (cancelDeployment, deleteDeployment) are destructive/reversible, but only the function name hints at this. Descriptions do not warn LLMs of side effects or suggest confirmation/dry-run patterns. An agent might call deleteDeployment without understanding it is irreversible.
Error handling returns error message strings only. No recovery guidance, no error classification (retryable vs fatal vs user-fixable), no actionable next steps. Example: catching errors and returning `Error getting deployment: <message>` tells the agent nothing about whether to retry, ask the user, or fail. Compare to production pattern: 'Deployment not found. Try list_deployments() to find available IDs.'
Parameter dependencies and mutual exclusivity not documented. Example: getDeployment accepts both 'deploymentId' (required) and 'slug' (optional). Are these alternative ways to identify the same deployment? Or does slug require teamId? Undocumented dependencies force LLMs to guess and risk malformed calls.
Pagination not documented. getDeploymentEvents and getDeployments accept 'limit' but no mention of offset/page, total_count, or has_more behavior. Large result sets risk blowing context windows. Production baseline requires pagination details in descriptions.
Numeric parameter ranges not constrained. 'limit' and 'from'/'to' timestamp parameters lack stated min/max values. LLMs might pass limit=999999 or timestamps in the wrong epoch, causing API errors or failures that handlers catch silently.
String enum parameters lack description clarity. Example: 'direction' has enum values ["forward", "backward"] but no explanation of which direction queries events (oldest first vs newest first?). LLMs might guess incorrectly and receive unexpected ordering.