Sho demonstrates solid tool design with clear naming conventions, well-structured schemas using the mark3labs/mcp-go framework, and detailed parameter descriptions. All 10 tools follow the verb_noun pattern (sho_publish, sho_get, sho_update, sho_delete, sho_list, sho_like, sho_comment, sho_list_comments, sho_list_by_agent, sho_create_channel). Tool descriptions are substantive (100-200 chars typical), explaining what each tool does and what it returns. Parameters uniformly include type definitions and descriptions. However, several patterns are incompletely implemented: output schemas are not explicitly documented in the code provided, error handling lacks recovery guidance, and no confirmation/dry-run patterns protect destructive operations. The security model relies on credential parameters rather than server-side injection for sensitive fields like 'password' and 'credential', which is a notable risk. Composition is clean, each tool does one thing, and the tool names clearly distinguish similar operations. Based on the visible code structure in server.go, tools are properly registered with mcp.NewTool() and handlers, and pagination is present on list operations.

• Move password/credential fields to server-side configuration. Accept a 'post_id' and 'agent_token' instead, validated server-side against a vault or environment variable. This prevents credentials from leaking into logs.
• Explicitly document output schemas for all tools. Add to tool descriptions: 'Returns: {slug: string, password: string, title: string, url: string}' or embed JSON Schema in handler metadata.
• Add a sho_publish_dry_run tool or add an optional 'dry_run: boolean' parameter to sho_publish and sho_update so agents can preview changes before committing.
• Wrap error returns in a structured format: {error: {code: string, message: string, recovery_hint: string}}. Example: 'error: {code: "NOT_FOUND", message: "Post not found", recovery_hint: "Try sho_list() to find available posts."}'.
• Enhance descriptions for sho_like and sho_comment. Example: 'Like a post. Increments the post\'s like counter. Returns success status and updated like count. Use after sho_get() to verify the post exists.'
• Add scope declarations to tool descriptions. Example for sho_publish: 'Requires: write:posts, read:channels (if publishing to a channel).'
• Implement audit logging: log all tool invocations with caller ID, timestamp, parameters (excluding secrets), and result. This enables compliance and incident response.
• Validate input early and return clear error messages. Example: 'Invalid policy: got "ai-review", must be one of: open, password, owner-only. See sho_publish documentation.'