This server has critical gaps across naming, descriptions, parameter definitions, and output documentation. While tool names follow a consistent verb_noun pattern (zentao_login, zentao_get_user, etc.), the implementations reveal major issues: (1) descriptions are in Chinese and minimal (10-20 chars), far below the 34-char baseline for adequate context; (2) parameter schemas lack required type fields and constraints, e.g., zentao_login has 'account' and 'password' but no minLength, maxLength, or pattern constraints; (3) output schemas are completely undocumented, tools return JSON via mcp.NewToolResultText() with no stated shape, making it impossible for LLMs to plan downstream tool composition; (4) critical parameters like 'page' and 'limit' in zentao_list_users are marked as 'string' but should be integers with range constraints (1-100, 1-500); (5) no error guidance, errors are wrapped in fmt.Errorf() with generic messages like '获取用户失败' (get user failed) that do not tell the agent whether to retry, lookup alternatives, or abort; (6) no handling of sensitive data, zentao_login and zentao_set_token expose credentials as parameters, violating the secret-injection pattern; (7) many tools (zentao_create_program, zentao_update_program, zentao_create_user) accept required parameters but the input schema shows {} or minimal fields, suggesting the handler implementation is incomplete or the schema is stale. The code sample only shows 16 of the claimed 82 tools, the remaining 66 are not visible for evaluation, forcing conservative scoring on unverified definitions.

• Move credentials (account, password, token) from tool parameters to server-side environment variables or a secure vault. zentao_login should authenticate via internal config, not expose credentials in the LLM prompt. Update zentao_set_token to accept a token from an environment variable or secure endpoint instead of as a string parameter.
• Document output schemas for all tools. For every tool, add a 'returns' or 'response' section describing the JSON object structure. Example for zentao_get_user: 'Returns {id, name, email, role, created_at, ...}'. This enables LLMs to extract and chain data.
• Expand all descriptions to 50-200 characters. Current descriptions (10-20 chars) are too brief. Example: 'zentao_list_users' should be: 'List all Zentao users with pagination. Returns user ID, name, email, role, and status. Use page and limit to control results. Returns up to 100 items per page.'
• Enforce numeric types and ranges for pagination. Change 'page' and 'limit' from 'string' to 'integer'. Add constraints: page minimum 1, limit minimum 1 and maximum 100. Document the default (page=1, limit=20).
• Add parameter descriptions to all tools. For zentao_list_users, describe 'page' as 'Page number (1-indexed, default 1)' and 'limit' as 'Items per page (1-100, default 20)'.
• Implement recovery guidance in error messages. Instead of 'fmt.Errorf("获取用户失败: %w", err)', return actionable errors: 'User not found. Try zentao_list_users() with a partial name.' or 'Invalid token. Call zentao_login() first.'