Reference, version 2026.7.21

AI Video Enhancer Studio CLI and MCP reference

aivideoenhancer is the command line of AI Video Enhancer Studio, installed with the desktop app. aivideoenhancer serve runs its MCP server for AI agents. This page lists every command, option, exit code and MCP tool.

aivideoenhancer <command> [arguments] [options]
aivideoenhancer enhance tape.avi -o restored --preset tape_restoration

Generated from aivideoenhancer catalog, version 2026.7.21. Every command also has --help. Older spellings of some commands and options still work, so existing scripts keep running, but this page and --help leave them out. The same reference is on GitHub: cli.md and mcp-server.md.

Commands

Command	What it does
`aivideoenhancer batch <input>`	Enhance every video in a folder (enhance with folder filters).
`aivideoenhancer catalog`	Print every CLI command and MCP tool as JSON (for AI agents, docs and the website).
`aivideoenhancer diagnose`	Generate a freeze-diagnostic zip (forensic log + Windows Event Log + GPU state).
`aivideoenhancer enhance <input>`	Enhance video files with AI.
`aivideoenhancer extract-frames <input> --output <output>`	Extract still frames from video(s) (optionally AI-upscale).
`aivideoenhancer formats`	List the video containers it reads and writes, the output codecs and the still formats.
`aivideoenhancer info <file>`	Show a video file's metadata: codec, resolution, duration, frame rate, bitrate, audio, HDR.
`aivideoenhancer presets`	List available enhancement presets.
`aivideoenhancer register <email> <key>`	Register your license key.
`aivideoenhancer remix`	Suggest and render highlight-reel remixes from a video.
`aivideoenhancer remix render <input> --clips <clips> --output <output>`	Render clip ranges into one H.264 MP4.
`aivideoenhancer remix suggest <input>`	Suggest highlight clip ranges via scene detection.
`aivideoenhancer serve`	Start MCP server for AI agent integration.
`aivideoenhancer status`	Show system status, GPU capabilities and license / trial state.
`aivideoenhancer watch <input>`	Watch a folder and enhance each video that arrives (until Ctrl+C).

Output options

These options work the same way in the four ContentaSoft command lines. catalog and serve take none of them: catalog always prints JSON and serve keeps stdout for the MCP protocol.

Option	Description
`--json`	Print one JSON object on stdout (success, exitCode, error, warnings, errors, outputs + the command's fields); nothing else goes to stdout.
`--plain`	Plain text: no colours, no escape codes, ASCII tables. Automatic when stdout is a pipe or a file, or when NO_COLOR is set.
`--quiet`	Print errors only.
`--verbose`	Verbose output: details and, on an error, the stack trace.
`--progress`	With --json: before the result object, one line per finished item, {"type":"progress","file":NAME,"completed":N,"total":T}.

Exit codes

The same table in the four ContentaSoft command lines. --help ends with it and catalog lists it.

Code	Name	Meaning
0	`ok`	OK: everything asked for was done.
1	`unexpected_error`	Unexpected error (a bug or a broken install); see the log.
2	`invalid_arguments`	Invalid arguments: unknown option, missing or malformed value; nothing was done.
3	`input_not_found`	Input not found or unreadable; nothing was done.
4	`partial_failure`	Partial failure: some items succeeded, some failed.
5	`all_failed`	All failed: no item succeeded.
6	`license_refused`	License or trial refused the run.
130	`cancelled`	Cancelled (Ctrl+C).

JSON output

With --json a command prints one JSON object on stdout and nothing else; logs go to stderr. Every object has these keys, beside the fields of the command itself:

Key	Meaning
`success`	`true` when the run did everything it was asked to do
`exitCode`	the process exit code (see Exit codes)
`error`	one sentence when the run failed, otherwise `null`
`warnings`	notes about the run that did not stop it
`errors`	one entry per item that failed (a batch keeps going)
`outputs`	the files the run wrote

Trial keys, present only when they apply:

Key	Meaning
`trialWatermarked`	the output carries the trial watermark
`trialExport`	the output is a trial export
`trialNotice`	one sentence about the trial, for the user
`freeFilesRemaining`	free full-quality files left on this computer
`buyUrl`	where to buy a licence

With --json --progress a long command first prints one progress line per finished item (NDJSON), then the result object last. A script that wants only the result skips the lines that start with {"type":"progress". Progress keys:

Key	Meaning
`type`	always `"progress"`
`file`	the file that finished
`completed`	items finished so far
`total`	items in the run

Command reference

`aivideoenhancer batch`

Enhance every video in a folder (enhance with folder filters).

aivideoenhancer batch <input> [options]

Argument	Description
`<input>`	Folder of videos to enhance.

Option	Type	Description
`--output`	string	Output directory (default: beside each source).
`--preset`	string	Preset name (e.g. tape_restoration for VHS/Hi8/MiniDV captures: CPU only, no graphics card needed; `presets` lists them all). With the other presets an interlaced source (DV, DVD, TV capture) gets --deinterlace yadif automatically unless you pass --deinterlace yourself.
`--profile`	string	Load the enhancement settings from a JSON file (the EnhancementProfile fields, e.g. {"upscale":"X2","denoise":"Light","outputCodec":"H265","crfValue":18}); the options below override it. Not with --preset.
`--upscale`	string	Upscale: off\|enhance\|x2\|x3\|x4 (enhance = clean up at same resolution).
`--denoise`	string	Denoise: off\|light\|medium\|strong.
`--denoise-method`	string	Denoise method: nlmeans\|hqdn3d\|chroma (default: hqdn3d; chroma = colour noise reduced much harder than luma, for tape).
`--stabilize`	string	Stabilize: off\|light\|medium\|strong.
`--deinterlace`	string	Deinterlace: off\|yadif\|yadifbob\|bwdif\|bwdifbob (bwdifbob = full motion: one frame per field, 25i becomes 50p, 29.97i becomes 59.94p). With --preset and no --deinterlace, interlaced sources get yadif; --deinterlace off keeps it off.
`--sharpen`	string	Sharpen: off\|light\|medium\|strong.
`--rolling-shutter`	string	Rolling-shutter (jello) correction: off\|light\|medium\|strong.
`--interpolate`	string	Target frame rate for smoother motion (RIFE): off\|30\|60. Only raises the rate — a clip already at/above the target is left unchanged.
`--codec`	string	Output codec: h264\|h265\|av1\|vp9.
`--crf`	integer	Quality (CRF value, 1-51, lower=better).
`--upscale-tile`	integer	Upscale tile size override (0=auto). Accepted for script compatibility; NVIDIA Video Super Resolution has no tiles and ignores it.
`--frame-batch`	integer	Frames per AI-runner invocation (0=auto, the default: sized from free VRAM).
`--temp-dir`	string	Scratch folder for frame extraction / interpolation temp data (default: the desktop app's Settings value, or system temp).
`--skip-existing`	boolean	Leave a video alone when its output file already exists (default: write a renamed copy beside it). A skipped video is not enhanced and costs no free file.
`--overwrite`	boolean	Replace an output file that already exists (default: write a renamed copy beside it). Not with --skip-existing.
`--compare`	boolean	Also write a side-by-side before/after video (source left, enhanced right, labelled underneath) next to each output as <name>_compare.mp4.
`--compare-labels`	string	Custom labels for --compare, as "FIRST\|SECOND" (default: "BEFORE\|AFTER").
`--compare-layout`	string	Layout for --compare: horizontal\|vertical. horizontal (default) = side by side; vertical = stacked top/bottom, which avoids the very wide 32:9 frame two 16:9 panels make side by side.
`--compare-seconds`	number	Length of the --compare video in seconds, taken a third of the way into the video (default: the whole video; 30 gives a short clip to show a customer).
`--hum`	string	Remove mains hum from the audio: off\|auto\|50\|60 (notches at the mains frequency and its harmonics; auto = 50 Hz for a 25/50 fps picture, 60 Hz for 29.97/59.94). Used with a preset or another enhancement option; the audio is re-encoded.
`--crop-bottom`	boolean	With --preset tape_restoration: crop the head-switching noise band at the bottom of a VHS/Video8/Hi8 capture (12 lines of 576, 8 of 480). The picture keeps its proportions, it is not stretched to fill.
`--stabilize-keep`	string	With --stabilize (or a preset that stabilizes): keep at least this percent of the picture, 50-100, or auto (default). Stabilizing zooms in to hide the moving edge, by as much as the shake needs; a limit stops the zoom there and leaves the part of the movement it has no room for. 100 = no zoom at all, the edge of the picture then moves.
`--hdr`	string	HDR10 / HLG sources: keep\|sdr (default: keep). keep = the output stays 10-bit HDR when the job has no AI stage, no stabilization, no rolling-shutter correction and no NLMeans denoise, and the codec is h265, av1 or vp9; otherwise the picture is tone-mapped to SDR and a line says so. sdr = always tone-map to SDR (BT.709).
`--upscale-engine`	string	What upscales x2\|x3\|x4: auto\|rtx\|standard. auto (default) = NVIDIA RTX Video Super Resolution where it can run, otherwise the standard upscaler (Lanczos resize + sharpening on the CPU, not AI; any PC). The engine is printed before the job starts.
`--include`	string[]	Only videos matching this pattern (e.g. *.mp4, or mp4;mov); repeat the option or separate patterns with ';' or ','.
`--exclude`	string[]	Leave out videos whose name matches (same pattern forms as --include).
`--recursive`	boolean	Include subfolders (the output folder is never searched when it is inside the input).
`--workers`	integer	Videos enhanced at the same time (default 1; AI stages share one GPU, so more rarely helps).
`--report`	string	Write a CSV report, one row per video (input, status, output, trial_watermarked, error), to this file.

Output options: --json, --plain, --quiet, --verbose, --progress.

`aivideoenhancer catalog`

Print every CLI command and MCP tool as JSON (for AI agents, docs and the website).

aivideoenhancer catalog

`aivideoenhancer diagnose`

Generate a freeze-diagnostic zip (forensic log + Windows Event Log + GPU state).

aivideoenhancer diagnose

Output options: --json, --plain, --quiet, --verbose, --progress.

`aivideoenhancer enhance`

Enhance video files with AI.

aivideoenhancer enhance <input> [options]

Argument	Description
`<input>`	Input video file or directory (a directory is searched with its subfolders; `batch` has the folder filters).

Option	Type	Description
`--output`	string	Output directory (default: beside each source).
`--preset`	string	Preset name (e.g. tape_restoration for VHS/Hi8/MiniDV captures: CPU only, no graphics card needed; `presets` lists them all). With the other presets an interlaced source (DV, DVD, TV capture) gets --deinterlace yadif automatically unless you pass --deinterlace yourself.
`--profile`	string	Load the enhancement settings from a JSON file (the EnhancementProfile fields, e.g. {"upscale":"X2","denoise":"Light","outputCodec":"H265","crfValue":18}); the options below override it. Not with --preset.
`--upscale`	string	Upscale: off\|enhance\|x2\|x3\|x4 (enhance = clean up at same resolution).
`--denoise`	string	Denoise: off\|light\|medium\|strong.
`--denoise-method`	string	Denoise method: nlmeans\|hqdn3d\|chroma (default: hqdn3d; chroma = colour noise reduced much harder than luma, for tape).
`--stabilize`	string	Stabilize: off\|light\|medium\|strong.
`--deinterlace`	string	Deinterlace: off\|yadif\|yadifbob\|bwdif\|bwdifbob (bwdifbob = full motion: one frame per field, 25i becomes 50p, 29.97i becomes 59.94p). With --preset and no --deinterlace, interlaced sources get yadif; --deinterlace off keeps it off.
`--sharpen`	string	Sharpen: off\|light\|medium\|strong.
`--rolling-shutter`	string	Rolling-shutter (jello) correction: off\|light\|medium\|strong.
`--interpolate`	string	Target frame rate for smoother motion (RIFE): off\|30\|60. Only raises the rate — a clip already at/above the target is left unchanged.
`--codec`	string	Output codec: h264\|h265\|av1\|vp9.
`--crf`	integer	Quality (CRF value, 1-51, lower=better).
`--upscale-tile`	integer	Upscale tile size override (0=auto). Accepted for script compatibility; NVIDIA Video Super Resolution has no tiles and ignores it.
`--frame-batch`	integer	Frames per AI-runner invocation (0=auto, the default: sized from free VRAM).
`--temp-dir`	string	Scratch folder for frame extraction / interpolation temp data (default: the desktop app's Settings value, or system temp).
`--skip-existing`	boolean	Leave a video alone when its output file already exists (default: write a renamed copy beside it). A skipped video is not enhanced and costs no free file.
`--overwrite`	boolean	Replace an output file that already exists (default: write a renamed copy beside it). Not with --skip-existing.
`--compare`	boolean	Also write a side-by-side before/after video (source left, enhanced right, labelled underneath) next to each output as <name>_compare.mp4.
`--compare-labels`	string	Custom labels for --compare, as "FIRST\|SECOND" (default: "BEFORE\|AFTER").
`--compare-layout`	string	Layout for --compare: horizontal\|vertical. horizontal (default) = side by side; vertical = stacked top/bottom, which avoids the very wide 32:9 frame two 16:9 panels make side by side.
`--compare-seconds`	number	Length of the --compare video in seconds, taken a third of the way into the video (default: the whole video; 30 gives a short clip to show a customer).
`--clip`	string	Export only a 10-second clip starting at this time (seconds, or m:ss / h:mm:ss) of ONE input file, named <name>-clip. On the trial a clip is clean and full resolution and does not use a free export; clips from one video can cover up to 30 seconds a day.
`--hum`	string	Remove mains hum from the audio: off\|auto\|50\|60 (notches at the mains frequency and its harmonics; auto = 50 Hz for a 25/50 fps picture, 60 Hz for 29.97/59.94). Used with a preset or another enhancement option; the audio is re-encoded.
`--crop-bottom`	boolean	With --preset tape_restoration: crop the head-switching noise band at the bottom of a VHS/Video8/Hi8 capture (12 lines of 576, 8 of 480). The picture keeps its proportions, it is not stretched to fill.
`--stabilize-keep`	string	With --stabilize (or a preset that stabilizes): keep at least this percent of the picture, 50-100, or auto (default). Stabilizing zooms in to hide the moving edge, by as much as the shake needs; a limit stops the zoom there and leaves the part of the movement it has no room for. 100 = no zoom at all, the edge of the picture then moves.
`--hdr`	string	HDR10 / HLG sources: keep\|sdr (default: keep). keep = the output stays 10-bit HDR when the job has no AI stage, no stabilization, no rolling-shutter correction and no NLMeans denoise, and the codec is h265, av1 or vp9; otherwise the picture is tone-mapped to SDR and a line says so. sdr = always tone-map to SDR (BT.709).
`--upscale-engine`	string	What upscales x2\|x3\|x4: auto\|rtx\|standard. auto (default) = NVIDIA RTX Video Super Resolution where it can run, otherwise the standard upscaler (Lanczos resize + sharpening on the CPU, not AI; any PC). The engine is printed before the job starts.

Output options: --json, --plain, --quiet, --verbose, --progress.

`aivideoenhancer extract-frames`

Extract still frames from video(s) (optionally AI-upscale).

aivideoenhancer extract-frames <input> --output <output> [options]

Argument	Description
`<input>`	Input video file or directory.

Option	Type	Description
`--output`	string, required	Output root directory (per-video subfolders created).
`--preset`	string	Sampling preset: storyboard\|thumbnails\|social\|scenes\|custom (scenes = one frame per detected scene change).
`--count`	integer	Evenly spaced frame count (implies custom even strategy); with --preset scenes it caps the top-N scenes by score.
`--timestamps`	string	Comma-separated seek positions in seconds (the same positions for every video).
`--upscale`	string	Upscale: off\|enhance\|x2\|x3\|x4 (default: off).
`--auto-upscale-below`	integer	If >0 and --upscale off, use x2 when height is below this (0=off).
`--format`	enum	Still format: png (lossless, default), jpg or webp.
`--quality`	integer	Lossy quality 1-100 for jpg/webp (default 90; ignored for png).
`--sharpest`	boolean	Per frame, compare 5 nearby candidates (+/-0.4s, +/-0.2s, exact) by Laplacian-variance sharpness and keep the sharpest (~5x extra small seeks per frame).

Output options: --json, --plain, --quiet, --verbose, --progress.

`aivideoenhancer formats`

List the video containers it reads and writes, the output codecs and the still formats.

aivideoenhancer formats [options]

Option	Type	Description
`--input`	boolean	Show only what is read.
`--output`	boolean	Show only what is written.

Output options: --json, --plain, --quiet, --verbose, --progress.

`aivideoenhancer info`

Show a video file's metadata: codec, resolution, duration, frame rate, bitrate, audio, HDR.

aivideoenhancer info <file>

Argument	Description
`<file>`	Video file to inspect.

Output options: --json, --plain, --quiet, --verbose, --progress.

`aivideoenhancer presets`

List available enhancement presets.

aivideoenhancer presets

Output options: --json, --plain, --quiet, --verbose, --progress.

`aivideoenhancer register`

aivideoenhancer register <email> <key>

Argument	Description
`<email>`	Registration email.
`<key>`	License key.

Output options: --json, --plain, --quiet, --verbose, --progress.

`aivideoenhancer remix`

Suggest and render highlight-reel remixes from a video.

Subcommands: remix render, remix suggest.

`aivideoenhancer remix render`

Render clip ranges into one H.264 MP4.

aivideoenhancer remix render <input> --clips <clips> --output <output> [options]

Argument	Description
`<input>`	Input video file.

Option	Type	Description
`--clips`	string, required	Clip ranges: "start-end,start-end,..." (sec or hh:mm:ss).
`--output`	string, required	Output MP4 path.
`--max-duration`	number	Cap total output length (seconds).
`--mute`	boolean	Drop audio.
`--width`	integer	Scale output to this width (height auto, even).
`--aspect`	enum	Center-crop to an aspect ratio: original, 9:16 (vertical Shorts/Reels), 1:1 (square) or 16:9 (wide).
`--music`	string	Background music audio file (mp3/wav/m4a/flac); looped to cover the reel, 2s fade-out at the end.
`--music-volume`	integer	Music volume 0-100 (default 30).
`--duck`	boolean	Duck the music under the original audio (sidechain compression); needs source audio, ignored with --mute.

`aivideoenhancer remix suggest`

Suggest highlight clip ranges via scene detection.

aivideoenhancer remix suggest <input> [options]

Argument	Description
`<input>`	Input video file.

Option	Type	Description
`--duration`	number	Target total highlight duration in seconds.
`--clips`	integer	Number of clips (default derived from --duration, 5-8).
`--window-start`	string	Analysis window start (sec or hh:mm:ss); required for videos > 4 hours.
`--window-end`	string	Analysis window end (sec or hh:mm:ss); defaults to start + up to 4 hours.
`--min-clip`	number	Minimum clip length (seconds).
`--max-clip`	number	Maximum clip length (seconds).
`--threshold`	number	Scene-detection threshold 0-100 (higher = fewer cuts).
`--audio-weight`	number	Blend weight 0-1 for audio-loudness spikes vs scene changes (0 = scene detection only); finds loud moments in static-camera footage.

`aivideoenhancer serve`

Start MCP server for AI agent integration.

aivideoenhancer serve

`aivideoenhancer status`

Show system status, GPU capabilities and license / trial state.

aivideoenhancer status

Output options: --json, --plain, --quiet, --verbose, --progress.

`aivideoenhancer watch`

Watch a folder and enhance each video that arrives (until Ctrl+C).

aivideoenhancer watch <input> [options]

Argument	Description
`<input>`	Folder to watch.

Option	Type	Description
`--output`	string	Output directory (default: beside each source).
`--preset`	string	Preset name (e.g. tape_restoration for VHS/Hi8/MiniDV captures: CPU only, no graphics card needed; `presets` lists them all). With the other presets an interlaced source (DV, DVD, TV capture) gets --deinterlace yadif automatically unless you pass --deinterlace yourself.
`--profile`	string	Load the enhancement settings from a JSON file (the EnhancementProfile fields, e.g. {"upscale":"X2","denoise":"Light","outputCodec":"H265","crfValue":18}); the options below override it. Not with --preset.
`--upscale`	string	Upscale: off\|enhance\|x2\|x3\|x4 (enhance = clean up at same resolution).
`--denoise`	string	Denoise: off\|light\|medium\|strong.
`--denoise-method`	string	Denoise method: nlmeans\|hqdn3d\|chroma (default: hqdn3d; chroma = colour noise reduced much harder than luma, for tape).
`--stabilize`	string	Stabilize: off\|light\|medium\|strong.
`--deinterlace`	string	Deinterlace: off\|yadif\|yadifbob\|bwdif\|bwdifbob (bwdifbob = full motion: one frame per field, 25i becomes 50p, 29.97i becomes 59.94p). With --preset and no --deinterlace, interlaced sources get yadif; --deinterlace off keeps it off.
`--sharpen`	string	Sharpen: off\|light\|medium\|strong.
`--rolling-shutter`	string	Rolling-shutter (jello) correction: off\|light\|medium\|strong.
`--interpolate`	string	Target frame rate for smoother motion (RIFE): off\|30\|60. Only raises the rate — a clip already at/above the target is left unchanged.
`--codec`	string	Output codec: h264\|h265\|av1\|vp9.
`--crf`	integer	Quality (CRF value, 1-51, lower=better).
`--upscale-tile`	integer	Upscale tile size override (0=auto). Accepted for script compatibility; NVIDIA Video Super Resolution has no tiles and ignores it.
`--frame-batch`	integer	Frames per AI-runner invocation (0=auto, the default: sized from free VRAM).
`--temp-dir`	string	Scratch folder for frame extraction / interpolation temp data (default: the desktop app's Settings value, or system temp).
`--skip-existing`	boolean	Leave a video alone when its output file already exists (default: write a renamed copy beside it). A skipped video is not enhanced and costs no free file.
`--overwrite`	boolean	Replace an output file that already exists (default: write a renamed copy beside it). Not with --skip-existing.
`--compare`	boolean	Also write a side-by-side before/after video (source left, enhanced right, labelled underneath) next to each output as <name>_compare.mp4.
`--compare-labels`	string	Custom labels for --compare, as "FIRST\|SECOND" (default: "BEFORE\|AFTER").
`--compare-layout`	string	Layout for --compare: horizontal\|vertical. horizontal (default) = side by side; vertical = stacked top/bottom, which avoids the very wide 32:9 frame two 16:9 panels make side by side.
`--compare-seconds`	number	Length of the --compare video in seconds, taken a third of the way into the video (default: the whole video; 30 gives a short clip to show a customer).
`--hum`	string	Remove mains hum from the audio: off\|auto\|50\|60 (notches at the mains frequency and its harmonics; auto = 50 Hz for a 25/50 fps picture, 60 Hz for 29.97/59.94). Used with a preset or another enhancement option; the audio is re-encoded.
`--crop-bottom`	boolean	With --preset tape_restoration: crop the head-switching noise band at the bottom of a VHS/Video8/Hi8 capture (12 lines of 576, 8 of 480). The picture keeps its proportions, it is not stretched to fill.
`--stabilize-keep`	string	With --stabilize (or a preset that stabilizes): keep at least this percent of the picture, 50-100, or auto (default). Stabilizing zooms in to hide the moving edge, by as much as the shake needs; a limit stops the zoom there and leaves the part of the movement it has no room for. 100 = no zoom at all, the edge of the picture then moves.
`--hdr`	string	HDR10 / HLG sources: keep\|sdr (default: keep). keep = the output stays 10-bit HDR when the job has no AI stage, no stabilization, no rolling-shutter correction and no NLMeans denoise, and the codec is h265, av1 or vp9; otherwise the picture is tone-mapped to SDR and a line says so. sdr = always tone-map to SDR (BT.709).
`--upscale-engine`	string	What upscales x2\|x3\|x4: auto\|rtx\|standard. auto (default) = NVIDIA RTX Video Super Resolution where it can run, otherwise the standard upscaler (Lanczos resize + sharpening on the CPU, not AI; any PC). The engine is printed before the job starts.
`--include`	string[]	Only videos matching this pattern (e.g. *.mp4, or mp4;mov); repeat the option or separate patterns with ';' or ','.
`--exclude`	string[]	Leave out videos whose name matches (same pattern forms as --include).
`--recursive`	boolean	Watch subfolders too (the output folder is never watched when it is inside the input).
`--interval`	integer	Seconds between two looks at the folder (default 10, at least 2). A new file is taken once it held still for one interval.
`--report`	string	When the watch stops, write a CSV report, one row per video (input, status, output, trial_watermarked, error), to this file.

Output options: --json, --plain, --quiet, --verbose, --progress.

MCP server

aivideoenhancer serve exposes 6 tools over the Model Context Protocol (stdio, newline-delimited JSON-RPC 2.0; protocol versions 2025-11-25, 2025-06-18, 2025-03-26 and 2024-11-05), so an AI agent such as Claude Code, Claude Desktop, Cursor or Codex can work on files on your PC. Server name: ai-video-enhancer. Pass absolute paths to every tool.

{
  "mcpServers": {
    "ai-video-enhancer": {
      "command": "aivideoenhancer",
      "args": [
        "serve"
      ]
    }
  }
}

In Claude Code one command adds it: claude mcp add ai-video-enhancer -- aivideoenhancer serve. Stdout carries JSON-RPC only, ping is answered while a tool runs, notifications/cancelled stops a running call, and a tool that cannot do its job returns isError: true with a sentence that names the problem.

Tool	Title	Read-only	Destructive	Idempotent	Network
`enhance_video`	Enhance video	no	no	no	no
`get_file_info`	Get video file info	yes	no	yes	no
`get_license_status`	Get license and trial status	yes	no	yes	no
`get_status`	Get GPU and license status	yes	no	yes	no
`list_formats`	List formats	yes	no	yes	no
`list_presets`	List presets	yes	no	yes	no

Destructive: the tool can replace a file that already exists. Network: the tool uses the internet.

`enhance_video`

writes files

Enhance video. Enhance a video using AI upscaling, frame interpolation, stabilization, and denoising.

Parameter	Type	Required	Description
`input_path`	string	yes	Absolute path to the input video file.
`output_path`	string	no	Output file path (ending .mp4, .mov or .mkv) or output directory; a relative path is resolved from the server's working directory (default: same directory as input).
`preset`	string	no	Preset ID: tape_restoration (VHS, Video8, Hi8 and MiniDV captures: full-motion deinterlace to 50p/59.94p, colour noise and hum removal, square-pixel 4:3; CPU only, no graphics card needed), old_video_restoration, surveillance_enhancement, content_creation, drone_action_cam, animation_anime, video_archival, enhance_cleanup, smooth_motion. list_presets says what each one needs from the PC.
`upscale`	string	no	Upscale mode: off, enhance (same-resolution cleanup), x2, x3, x4. One of: `off`, `enhance`, `x2`, `x3`, `x4`.
`upscale_engine`	string	no	What upscales x2, x3, x4: auto (default: NVIDIA RTX Video Super Resolution where it can run, otherwise the standard upscaler), rtx, standard (Lanczos resize + sharpening on the CPU, not AI; any PC). The result carries upscaleEngine. One of: `auto`, `rtx`, `standard`.
`denoise`	string	no	Denoise level: off, light, medium, strong. One of: `off`, `light`, `medium`, `strong`.
`stabilize`	string	no	Stabilization level: off, light, medium, strong. One of: `off`, `light`, `medium`, `strong`.
`stabilize_keep`	integer	no	With stabilize (or a preset that stabilizes): keep at least this percent of the picture, 50-100. Stabilizing zooms in to hide the moving edge, by as much as the shake needs; a limit stops the zoom there and leaves the part of the movement it has no room for. 100 = no zoom at all, the edge of the picture then moves. Leave it out to let the stabilizer choose. Range: 50 to 100.
`deinterlace`	string	no	Deinterlace: off, yadif, yadifbob, bwdif, bwdifbob (full motion: one frame per field, 25i becomes 50p). With a preset and no deinterlace value, an interlaced source (DV, DVD, TV capture) gets yadif automatically; pass off to keep it off. One of: `off`, `yadif`, `yadifbob`, `bwdif`, `bwdifbob`.
`denoise_method`	string	no	Denoise method: hqdn3d (default), nlmeans, chroma (colour noise reduced much harder than luma, for tape). One of: `hqdn3d`, `nlmeans`, `chroma`.
`hum`	string	no	Remove mains hum from the audio: off, auto (50 Hz for a 25/50 fps picture, 60 Hz for 29.97/59.94), 50, 60. Used with a preset or another enhancement; the audio is re-encoded. One of: `off`, `auto`, `50`, `60`.
`crop_bottom`	boolean	no	With preset tape_restoration: crop the head-switching noise band at the bottom of a VHS/Video8/Hi8 capture (12 lines of 576, 8 of 480); the picture keeps its proportions.
`sharpen`	string	no	Sharpen: off, light, medium, strong. One of: `off`, `light`, `medium`, `strong`.
`rolling_shutter`	string	no	Rolling-shutter (jello) correction: off, light, medium, strong. One of: `off`, `light`, `medium`, `strong`.
`interpolate`	string	no	Target frame rate for smoother motion (RIFE): off, 30, 60. Only raises the rate — a clip already at/above the target is left unchanged. One of: `off`, `30`, `60`.
`codec`	string	no	Output codec: h264, h265, av1, vp9. One of: `h264`, `h265`, `av1`, `vp9`.
`hdr`	string	no	HDR10 / HLG sources: keep (default) or sdr. keep = the output stays 10-bit HDR when the job has no AI stage (upscale, interpolate), no stabilize, no rolling_shutter and no nlmeans denoise, and the codec is h265, av1 or vp9; otherwise the picture is tone-mapped to SDR. sdr = always tone-map to SDR (BT.709). hdrNotice in the result says what was done. One of: `keep`, `sdr`.
`crf`	integer	no	Quality (CRF value, 1-51, lower = better quality, default: 18). Range: 1 to 51.
`skip_existing`	boolean	no	Leave the video alone when its output file already exists (default: write a renamed copy beside it). A skipped video is not enhanced and costs no free file.
`overwrite`	boolean	no	Replace the output file when it already exists (default: write a renamed copy beside it). Not with skip_existing.
`clip`	string	no	Export only a 10-second clip starting at this time (seconds such as 42 or 42.5, or m:ss / h:mm:ss such as 1:05) instead of the whole video, named <name>-clip; the result carries clipPath. On the trial a clip is clean and full resolution and does not use a free export; clips from one video can cover up to 30 seconds a day.
`compare`	boolean	no	Also write a side-by-side before/after video (source and enhanced, labelled) next to each output as <name>_compare.mp4; the result carries comparePaths.
`compare_labels`	string	no	Custom labels for compare, as "FIRST\|SECOND" (default: "BEFORE\|AFTER").
`compare_seconds`	number	no	Length of the compare video in seconds, taken a third of the way into the video (default 0: the whole video; 30 gives a short clip to show a customer). Range: 0 to 3600.
`compare_layout`	string	no	Layout for compare: horizontal (default, side by side) or vertical (stacked top/bottom, avoids the very wide 32:9 frame two 16:9 panels make side by side). One of: `horizontal`, `vertical`.

`get_file_info`

read-onlyidempotent

Get video file info. Get a video file's metadata: codec, resolution, duration, bitrate, frame rate, audio, HDR (and what enhance_video will do with the HDR).

Parameter	Type	Required	Description
`input_path`	string	yes	Absolute path to the video file.

`get_license_status`

read-onlyidempotent

Get license and trial status. Get the license and trial state before enhancing: registered or trial, free full exports left (freeFilesRemaining), and what exports become after the free ones (afterFreeFiles: watermark_720p = a watermark and output capped at 1280x720; 10-second clips are never branded). nextOutput says what the next full export will be: clean or watermark_720p.

No parameters.

`get_status`

read-onlyidempotent

Get GPU and license status. Get system status including GPU capabilities, available tools, and license state.

No parameters.

`list_formats`

read-onlyidempotent

List formats. List the video containers AI Video Enhancer reads and writes, the output codecs (enhance_video codec) and the still formats of frame extraction: format, extensions, canImport, canExport, category.

Parameter	Type	Required	Description
`input_only`	boolean	no	Only what is read (the CLI's --input).
`output_only`	boolean	no	Only what is written (the CLI's --output).

`list_presets`

read-onlyidempotent

List presets. List available video enhancement presets.

No parameters.