The server registers 20 tools with explicitly visible schemas and descriptions in src/tools/server.ts and src/tools/automod.ts. Most tools have moderate-quality descriptions (100-250 chars), proper naming with action verbs, and JSON Schema input definitions. However, several critical gaps prevent a higher score: (1) output schemas are undocumented, tools lack explicit response field definitions, forcing LLMs to infer structure; (2) parameter descriptions are present but often generic, lacking format/range constraints that would guide LLM input validation (e.g., 'limit' has no min/max stated); (3) error handling guidance is absent, tools provide no recovery hints; (4) some parameter relationships are undocumented (e.g., create_automod_rule has 4 mutually exclusive trigger types but no explicit 'choose exactly one' language); (5) many parameters accept IDs but do not document whether fuzzy name matching applies (only 'channel' and 'emoji' explicitly note fuzzy matching). Per-tool scores average 56 - 62, clustering in the fair-to-good range. This is typical of community-maintained MCP servers that prioritize breadth (20 tools) over depth (complete, LLM-optimized schemas).

• Document output schema for every tool. Use a consistent format: 'Returns: {type: object, properties: {field_name: {type: ..., description: ...}, ...}}'. Example for list_invites: 'Returns: {invites: [{code: string, uses: number, max_uses: number, expires_at: string|null, created_at: string}, ...], total: number}'
• Add min/max constraints to numeric parameters. Change 'limit' descriptions to include bounds: 'Number of entries to fetch (1 - 100, default: 25)'. Update schema to include 'minimum' and 'maximum' fields.
• Explicitly document parameter relationships in descriptions. For create_automod_rule, add a note: 'Exactly one of triggerType must be chosen. If triggerType=keyword, provide keywords or regexPatterns. If triggerType=keyword_preset, provide presets. If triggerType=spam or mention_spam, do not provide keywords/regexPatterns/presets.'
• Standardize fuzzy name matching documentation. For automod tools, clarify: 'exemptRoles accepts role IDs or role names (fuzzy matched). exemptChannels accepts channel IDs or channel names (fuzzy matched).' This matches the pattern used in create_invite.
• Add error handling guidance to destructive tools. Example for delete_invite: 'If the invite code is invalid or already deleted, returns error "invite_not_found". To see active invites, call list_invites().'
• Implement pagination for list tools. Add limit (default 20, max 100) and offset/cursor parameters. Return total count or next_cursor in response. Document: 'Returns: {items: [...], total: number, hasMore: boolean}' (or cursor variant).