Two tools with minimal descriptions and incomplete schemas. list_endpoints has empty input schema (no properties defined), call_endpoint has basic parameter descriptions but lacks critical details on valid namespaces/endpoints, error cases, and output structure. Descriptions are under 100 chars and lack WHEN/WHY guidance. No parameter type constraints (enums for namespace), no output schema documentation, no error recovery guidance. Tool names are reasonable (verb_noun pattern) but lack the specificity needed for LLM disambiguation.

• Add enum constraint to call_endpoint.namespace with all valid values (bisim, eshot, otopark, egitim, saglik, kutuphane, etc.). Extract from src/endpoints/ directory.
• Document call_endpoint output schema: what fields does the API response contain? Return type (object/array)? Include example response structure.
• Expand tool descriptions to 120 - 180 chars. For list_endpoints: 'Retrieve all available Izmir Open Data API services and their methods. Call this first to discover which endpoints are available before invoking call_endpoint.' For call_endpoint: 'Invoke a specific Izmir Open Data API method. Requires namespace (service name) and endpoint_name (method). Returns raw API response. Use list_endpoints to discover available services.'
• Define call_endpoint.args item schema. If args are positional strings, use {"type": "string"}. If mixed types, use {"type": ["string", "number", "boolean"]}. Document in description: 'Array of arguments passed to the endpoint method, in order. Type and count depend on the specific endpoint, see list_endpoints for details.'
• Add error recovery guidance: 'If call_endpoint returns an error, verify namespace and endpoint_name exist by calling list_endpoints. If args are invalid, check the endpoint signature in list_endpoints output.'