Loading ToolBench
Social Media Manager MCP. The operational layer for running accounts: content calendars, posting cadence, caption frameworks, hashtag strategy, engagement playbooks, metric interpretation, and client reporting.
This server defines 7 tools with action-verb naming (get_*, build_*) and reasonable descriptions (avg ~150 chars). However, critical gaps undermine quality: most tools have NO input schema validation visible in the code, parameters are declared but lack type constraints, min/max bounds, or format specifications. Output schemas are completely undocumented; the code returns hardcoded objects (CADENCE, CALENDAR_SYSTEM, etc.) but never specifies what fields agents should expect. Error handling is absent, no recovery guidance, no validation errors, no actionable failure messages. Parameters like 'platform' and 'metric' use enums in the schema but lack descriptions explaining when/why to use each value. The server is read-only (no state mutation), which is good for safety, but the lack of structured output documentation and input validation makes it difficult for LLMs to compose these tools reliably.
Client reporting structure: headline (one number that moved), what we posted (by pillar), the 3 metrics that matter for the goal, one learning + one change, and next period plan. Includes the rule (report on goal, not vanity) and cadence (monthly full + weekly pulse).
The content-calendar system: pillar-first assignment, a weekly shape (Mon authority → Fri engagement), the batch-by-task rule, and the conversion ratio. Use to plan a week/month for an account.
Four caption structures: hook-expand-cta (standard video), story (narrative), list (carousels/tips), question-seed (engagement). Each includes the structure, when to use, and the rule.
Community management and engagement tactics: first-hour replies, reply-to-reply threading, pinned comments, DM strategy, outbound engagement, and handling negativity.
Hashtag count, mix, and platform-specific notes for tiktok, instagram, linkedin, x, youtube.
Posting frequency, best windows, and the key note per platform: tiktok, instagram, linkedin, x, youtube. Omit platform for all.
Output schemas completely undocumented. Tools return hardcoded objects (CADENCE, CALENDAR_SYSTEM, CAPTION_FRAMEWORKS, etc.) but the code never declares what fields, types, or structure agents should expect. LLMs cannot plan downstream tool calls or extract data reliably without knowing the response shape.
Parameter descriptions missing or incomplete. Enum parameters like 'platform' and 'metric' lack descriptions explaining what each value means, when to use it, or what the tool returns for each choice. LLMs cannot disambiguate between similar enum values without explicit guidance.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 49 | 2026-07-28+ | v2 |
What each metric means (reach, engagement rate, saves, shares, watch-time retention, follower growth, profile visits to follow) and how to fix it if it's low.
No input validation or error handling. The code accepts parameters but never validates them (e.g., checking if 'platform' is in the enum, or if 'metric' is valid). No error responses guide the LLM on what to do if a parameter is invalid or the tool fails.
Tools with no parameters (build_content_calendar, get_engagement_playbook, build_client_report) return static data regardless of context. No way for agents to customize output (e.g., filter by account type, time period, or goal). Reduces composability and forces agents to post-process generic responses.
No pagination or result limits documented. Tools return entire objects (all platforms, all frameworks, all metrics) with no way to limit or paginate results. If these objects grow, responses could bloat the context window.