Three tools with basic schemas and descriptions, but significant gaps in parameter documentation and output structure. slack_list_channels has no input schema (empty object). slack_read_channel and slack_post_message have typed parameters with descriptions, but descriptions lack actionable constraints (e.g., no mention of channel ID format, no guidance on when to use name vs ID). No output schemas documented, LLMs cannot predict response structure. No error handling guidance. Tool names follow verb_noun pattern (good), but descriptions are generic and under-optimize for LLM selection. No tool annotations (readOnlyHint, destructiveHint) despite clear risk profiles.

• Add explicit output schema documentation for each tool. Example: slack_list_channels returns {type: 'string', description: 'Formatted list of channels with name, ID, topic, member count'}. slack_read_channel returns {type: 'string', description: 'Formatted messages with timestamp, user, text'}.
• Enhance parameter descriptions with format and constraint details. Example: 'channel: Slack channel name (without #, 1-80 chars, alphanumeric + hyphens) or channel ID (starts with C, e.g. C1234567890). If unsure, call slack_list_channels first.'
• Add tool annotations via FastMCP. Example: @mcp.tool(annotations={'readOnlyHint': True}) for slack_list_channels; @mcp.tool(annotations={'destructiveHint': True}) for slack_post_message.
• Implement error handling with recovery guidance. Example: 'Channel not found. Available channels: [list]. Try slack_list_channels() to discover valid channels.' For slack_post_message, return 'Message sent to #channel (ts=1234567890)' with timestamp for idempotency.
• Add input validation and return structured error messages. Example: if limit < 1 or limit > 100, return 'Invalid limit: {limit}. Must be 1-100.' instead of silent clamping.