Server defines 7 tools with basic schemas and descriptions, but significant gaps limit production readiness. All tools have descriptions (10-60 chars, mostly short), and input schemas are present with type definitions. However, parameter descriptions are sparse or missing entirely, output schemas are undocumented, and error handling lacks recovery guidance. No tool annotations (readOnlyHint/destructiveHint) despite clear risk classifications. Naming is verb-forward (jira_search, github_get_file) but inconsistent (pagerduty_ack vs pagerduty_incidents). WRITE tools (jira_create, pagerduty_ack) lack confirmation/dry-run patterns. Knowledge_search tool has minimal parameter documentation.

• Add output schema documentation to every tool. Specify return type (object/array), field names, types, and whether pagination is supported. Example: jira_search should document it returns {issues: [{key, summary, status, assignee}], total_count}.
• Expand parameter descriptions to 50-100 chars each. For jira_search.max_results, add 'Maximum number of results to return (1-100, default 10)'. For knowledge_search.team, specify valid team names or enum values.
• Add tool annotations: mark jira_create and pagerduty_ack with destructiveHint=true; mark all READ_ONLY tools with readOnlyHint=true. This enables clients to enforce safety policies.
• Implement confirmation pattern for WRITE tools. jira_create and pagerduty_ack should support a 'dry_run' parameter or return a confirmation token requiring explicit approval before execution.
• Document error scenarios and recovery paths in tool descriptions. E.g., 'If JQL is invalid, returns error with suggested syntax. If Jira is unreachable, returns timeout error, retry after 30s.' This guides agent retry logic.

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry