Server has 5 tools with explicit schemas and descriptions, but quality is inconsistent. Four tools (list_sources, get_source_prompts, list_amazon_vpc_lattice_prompts, get_amazon_vpc_lattice_prompts) are simple metadata/discovery tools with minimal descriptions (34-50 chars, below the 50-200 char LLM-optimized baseline). The critical vpc_lattice_cli tool accepts arbitrary AWS CLI commands via enum constraint, but lacks parameter descriptions for 'args' (generic 'Command arguments as key-value pairs' is insufficient), missing guidance on format, validation, or error recovery. No output schemas documented. Error handling exists but is minimal, McpError responses lack recovery guidance. Tool naming follows verb_noun convention (good), but descriptions are too terse to guide LLM selection effectively.

• Expand metadata tool descriptions to 80-150 chars. Example: 'list_sources' → 'List all available data sources with their URLs and associated sample prompts. Call this first to discover available sources before requesting prompts for a specific source.'
• Document vpc_lattice_cli 'args' parameter with explicit format guidance: 'Object with camelCase keys (e.g., {serviceName: "my-service"}). Keys are converted to kebab-case CLI flags. Refer to AWS VPC Lattice CLI docs for valid arguments per command.'
• Add output schema documentation for each tool. Example for list_sources: 'Returns array of {name: string, url: string, prompts: string[]}'. For vpc_lattice_cli: 'Returns parsed JSON output from AWS CLI (structure varies by command).'
• Enhance error messages with recovery hints. Example: 'Source not found: X. Available sources: [list]. Call list_sources() to see all options.' For AWS CLI errors: 'AWS CLI command failed: Y. Verify AWS credentials, region, and command syntax.'
• Sanitize 'args' object keys and values to prevent command injection. Validate keys against a whitelist of known VPC Lattice CLI arguments per command, and escape shell-unsafe characters in values.