No user-submitted related servers found.
Scored across 93 tools
Every tool's purpose is sharply delineated, and the descriptions actively name their sibling tools and state the difference (shot_sheet vs footage_sheet, reframe vs reframe_detect, transcribe vs attach_transcript). However, several sibling clusters — verify/finish_check/hold_check, the four reframe tools, the three sheet tools, and pack_apply/pack_activate — share names and mechanisms closely enough that an agent must read carefully to select correctly, and 93 tools compounds the selection pressure.
snake_case throughout with strong family conventions: cue_*/unspoken_*/hold_*/continuity_*/pack_*/review_* prefixes and consistent _add/_rm/_check/_detect/_apply suffixes. Deviations exist: list operations are inconsistently named (list_media vs cue_ls vs review_list vs describe_ls), and several single-noun get/set tools (head, tail, music, canvas, doctor, assets) break the verb_noun pattern.
At 93 tools this is far beyond the 25+ threshold and overwhelms the agent-facing surface, even though the domain is broad. The count forces every tool call through a huge selection space and makes the set impractical to navigate from a plain tool list. Consolidation — e.g., folding hold_*, pack_*, or review_* families together, or merging read-only reports — would be needed to bring it into a usable range.
The surface covers the full proofcut pipeline end-to-end: project lifecycle, import, transcription and transcript QA, timeline editing, cues, captions, cards, packs, framing, holds, music, export, render verification, continuity, and human review. CRUD is consistently closed per family (add/rm/ls for cues and unspoken marks, check/accept/reject/ls for continuity, apply/show/status for packs). No obvious dead ends: every mutation is undoable, and deliberately unsupported operations (like manual transcript correction) are explicitly designed around rather than omitted.