Loading ToolBench
MCP server for AutoWhisper — drive your AI CMO (generate, approve, and publish marketing content across 30+ platforms) from any MCP client.
AutoWhisper MCP exposes 13 tools with mixed quality. Tool names follow verb_noun convention (autowhisper_cmo, autowhisper_products, autowhisper_action), which is good. However, descriptions are inconsistent: some are terse (e.g., 'fast read-only product counts' for autowhisper_products_summary is 34 chars, below the 50-char baseline for clarity), while others lack actionable context. Input schemas are present for most tools but vary in completeness, several tools accept empty objects (autowhisper_products, autowhisper_status, autowhisper_feed, autowhisper_posts, autowhisper_wallet, autowhisper_platforms, autowhisper_connect) with no parameters, which is valid but offers no guidance. The autowhisper_cmo tool accepts a single 'instruction' string with minimal constraint documentation. Error handling is not visible in the provided code excerpt. Output schemas are not documented in the source. The server uses environment-based token injection (AUTOWHISPER_API_TOKEN, credentials file), which is correct for secret handling, but no per-tool permission scopes are declared.
deterministic feed/post actions
send an instruction, wait for the CMO's reply
approve/decline a high-impact action
device flow — get authorised without a copy-paste
deterministic field-level content edits
fast read-only CMO feed list
cross-channel results — did any of it work?
Seven read-only tools (products_summary, products, status, feed, posts, wallet, platforms) have descriptions under 35 characters, below the 50-char baseline for LLM clarity. Descriptions like 'fast read-only product counts' do not explain WHEN to call this tool vs. similar ones or what structure is returned.
Seven tools accept empty input schemas (no parameters). While valid, this provides no guidance to LLMs about optional filters, pagination, or scope. For example, autowhisper_products and autowhisper_feed should document whether they accept workspace_id, limit, or status filters.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 49 | 2026-07-28+ | v2 |
fast read-only connected platforms list
fast operational read of posts/delivery queue
fast read-only product list
fast read-only product counts
fast read-only account/CMO status
fast read-only wallet/credits balance
Output schemas are not documented in the source code. LLMs cannot plan downstream tool calls or extract required fields (e.g., product_id, feed_item_id) without knowing what each tool returns. This violates the pattern:tool requirement.
No error handling guidance visible in source. Tools like autowhisper_cmo (which polls for CMO reply) and autowhisper_connect (device flow) likely have failure modes (timeout, auth failure, network error) but no recovery hints are documented. LLMs cannot self-correct without actionable error messages.
No per-tool permission scopes declared. Tools like autowhisper_action and autowhisper_confirm are destructive (WRITE risk), but no scope metadata (e.g., 'write:feed', 'write:content') is visible. This prevents least-privilege agent configuration and audit clarity.