This HTTP MCP server exposes three read-only tools for Malloy semantic models. The tools themselves are straightforward, but definition quality is significantly hampered by: (1) minimal parameter descriptions for the 'query' tool; (2) missing output schemas across all tools; (3) no error handling guidance; (4) no documentation of what fields are returned or how results are paginated. The naming is acceptable (verb_noun pattern), but the schema and output documentation gaps are severe.

• Add complete output schemas for all three tools. Document the exact JSON structure, field names, types, and what each field represents. Use the pattern: {"type": "object", "properties": {...}, "required": [...]}.
• Expand tool descriptions to 100-150 characters with clear context: What does the tool reveal? When should it be called? Example: 'get_context: Retrieve the list of available Malloy semantic models and their data sources. Call this first to see what models are available before executing queries.'
• Add full parameter descriptions to 'query' tool. 'source' → 'Name of the Malloy semantic model to query (e.g. "sales_model").' 'query' → 'A valid Malloy query string. Syntax: {name | select field_name | filter condition | etc.}. See Malloy docs for full syntax.'
• Document error cases and recovery paths. Add to query tool: 'Errors: Returns 400 if source not found (call get_context() to list valid sources), 400 if query syntax is invalid (check syntax documentation), 504 if query timeout (simplify query or add time filters).'
• If results can be large, add page/offset and limit parameters to list_packages and query tools. Document the result limit: 'Returns up to 100 results; use offset/limit for pagination.'
• Link to Malloy query language docs so LLMs understand the query parameter format.