Server has 9 tools with complete input schemas and basic descriptions. However, descriptions are generic and lack actionable context for LLM selection. Parameter descriptions are minimal (e.g., 'The SOQL query to execute' without format guidance). No output schemas documented. Error handling is absent, no recovery guidance or categorization. Security concerns: credentials handled via environment variables (good), but no audit logging or permission gates visible. Tool composition is reasonable (single responsibility), but several tools expose dangerous operations (delete_record, apex_execute) without confirmation patterns. Naming follows verb_noun convention well. Overall: functional but below production quality.

• Document output schemas for all tools. Example for get_record: 'Returns {id: string, fields: {[key]: any}, attributes: {type: string, url: string}}'. This enables LLMs to chain results into downstream tools.
• Add confirmation/dry-run pattern to delete_record and apex_execute. Implement a 'confirm' parameter or separate 'preview_delete' tool to prevent accidental data loss.
• Enhance error messages with recovery guidance. Catch SalesforceError exceptions and return structured errors: {error: 'INVALID_SOQL', message: '...', suggestion: 'Check field names...', retryable: false}.
• Expand parameter descriptions with constraints and examples. For 'data' in create_record, add: 'Object with field names as keys (e.g., {Name: "Acme", Industry: "Tech"}). Use get_object_fields to discover valid fields for this object.'
• Add limit and offset parameters to run_soql_query and run_sosl_search. Cap default results at 20 and document: 'Returns up to 20 records. Use offset to paginate. Total count included in response.'