MCP server for video and audio processing using FFmpeg, supporting extraction, trimming, format conversion, overlays, and audio manipulation
VideoAudioServer demonstrates mediocre definition quality. Tool names follow verb_noun convention well (extract_audio_from_video, trim_video, convert_audio_properties, etc.), but descriptions are inconsistent in depth and quality. Parameter descriptions exist but are often generic. Critical gaps: (1) Most tools lack output schemas, the descriptions state what happens but not what structure is returned. (2) No error categorization or recovery guidance, errors are caught but responses are bare status strings with no actionable next steps for an LLM. (3) Several complex tools (convert_video_properties, add_text_overlay) have underdocumented parameter relationships and constraints. (4) No indication of which tools are idempotent or safe to retry. Average tool description length is 85 chars (well below the 194-char baseline); this brevity lacks sufficient context for LLM tool selection. Tool schemas show type declarations (string, integer, array) but lack constraint specifications (enums, regex patterns, min/max bounds).
Adds text overlays to a video with specified properties and timing.
Changes the aspect ratio of a video, using padding or cropping.
Converts audio file format and ALL specified properties like bitrate, sample rate, and channels.
Converts video file format and ALL specified properties like resolution, codecs, bitrates, and frame rate.
Extracts audio from a video file and saves it.
Returns a simple health status to confirm the server is running.
Trims a video to the specified start and end times.
No output schemas documented. Tools return free-text status strings (e.g., 'Audio extracted successfully to {output_audio_path}') rather than structured objects with typed fields. LLMs cannot reliably parse unstructured responses for downstream chaining or field extraction.
Error responses provide no recovery guidance. Catch blocks return raw ffmpeg.Error stderr or generic 'An unexpected error occurred' messages. LLMs cannot determine whether to retry, ask the user, or abandon the operation.
add_text_overlay parameter 'text_elements' is an array with no schema for array items. What fields does each text element require? Font, size, position, color, start_time, end_time? LLM must guess or fail.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 55 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 25 | - | v1 |
convert_video_properties and convert_audio_properties have many optional parameters with undocumented interdependencies. If 'resolution' is provided, does 'video_codec' matter? Can audio and video be converted independently or must both be specified? No guidance.
Parameter format constraints missing. 'start_time' and 'end_time' in trim_video accept 'HH:MM:SS or seconds' but no regex or enum clarifies valid formats. 'target_aspect_ratio' (e.g., '16:9', '4:3') has no validation. LLMs cannot reliably generate valid formats.
Tool descriptions are too brief (50 - 85 chars on average). Below the 194-char baseline, they lack sufficient context for LLM tool selection. Example: 'Converts audio file format and ALL specified properties like bitrate, sample rate, and channels.' does not explain WHEN to use this vs extract_audio_from_video or what the output looks like.
No indication of operation idempotence or mutability. Can trim_video be called twice with the same parameters and produce the same result? Is convert_audio_properties safe to retry? Agents need this information to manage retries and duplicate prevention.