Stellar MCP has 4 tools with basic verb-noun naming (get_*) and descriptions present. However, critical gaps undermine quality: (1) Most tools have EMPTY input schemas (get-tvl, get-volume, get-market-overview accept no parameters), making them non-composable and limiting agent flexibility. (2) Parameter descriptions are missing entirely, token_id in get-token-price has a description in the schema, but no guidance on valid formats or examples. (3) Output schemas are completely undocumented, responses are text blobs with no structured field definitions, forcing LLMs to parse unstructured text. (4) No error handling guidance, if a token_id is invalid, the tool returns a text message listing alternatives, but no structured error classification or recovery hints. (5) No pagination or result limits documented. This is a typical community server with functional tools but poor LLM integration.

• Document output schemas for all tools. Define structured response objects with typed fields (e.g., {price: number, currency: string, change_24h: number}) instead of text blobs. This enables LLMs to extract and chain data reliably.
• Add optional parameters to get-tvl, get-volume, get-market-overview: limit (1-100, default 20), sort_by (tvl|name), and offset for pagination. This gives agents control and prevents context bloat.
• Enhance token_id parameter description: 'Token identifier (lowercase, hyphen-separated). Examples: stellar (XLM), centre-usdc (USDC), ultrastellar-aqua (AQUA). Call get-available-tokens to discover valid IDs.'
• Implement structured error responses: {error: {type: 'not_found'|'invalid_input'|'rate_limit', message: '...', recovery: 'Try search_tokens() with partial name'}}. This guides agent retry logic.
• Add a discovery tool: get-available-tokens() → {tokens: [{id, name, symbol}]} to enable agents to explore valid token IDs before calling get-token-price.