Skip to main content
Glama

Server Details

VAR2 is a hosted creative MCP server for generating and editing images, creating videos, music, voiceovers and 3D assets, and assembling creative workflows. Connect with OAuth using a VAR2 account; generation uses account credits.

Ownership verified
Status
Healthy
Uptime
100.0% over the last 24 hours
Last Tested
Transport
Streamable HTTP · MCP 2025-06-18
URL
ChatGPT
–
Claude
–

TDQS

A3.8/5.0

Scored across 38 tools

Disambiguation3/5

The seven var2_get_*_result tools and the eight motion-graphics tools (create/preview/save/read/update/motion_plan/motion_build/motion_catalog) share heavy conceptual overlap and are separated mainly by long prose routing rules rather than by a self-evident purpose. Uploads are split across request_upload / upload_asset / confirm_upload, and clip editing across create_video / magic_replace / video_effect / video_look, which will cause frequent misselection. The descriptions do an unusually thorough job of disambiguating, which is what keeps this above a 2, but the boundaries are taught, not obvious.

Naming Consistency4/5

Almost everything follows var2_<verb>_<noun> in snake_case: create_*, get_*_result, save_*, list_models, trim_audio, join_videos. Minor deviations exist where the name is noun-shaped or verb-less (var2_magic_replace, var2_video_effect, var2_video_look, var2_motion_catalog), but the prefix and casing are uniform throughout.

Tool Count2/5

38 tools for a multi-modal generation platform is heavy; motion graphics alone consumes eight entries (plan, catalog, build, preview, save, read, update, create) that could be collapsed. Several tools are near-duplicates at the surface level (request_upload vs upload_asset, preview vs save vs create for motion), so not every tool clearly earns its place. The breadth of modalities explains part of the count but not the tool explosion within one sub-domain.

Completeness4/5

Coverage is strong across the lifecycle: create/get for image, video, audio, dialog, and 3D, plus upload, stitching, timeline rendering, voice catalog, character sheets, and cost estimation, with no obvious dead end for the core generate-then-fetch flow. Gaps remain around asset management — no delete/rename for library items, no cancel or list-jobs tool, and var2_get_results requires ids you already hold.

Publisher details

Operator
Var2 · Publisher source
Vendor relationship
First-party · Publisher source
Privacy policy
Not provided
Trust center
Not available
Restrictions
custom oauth and api key · Publisher source

Available Tools

38 tools
var2_check_join_statusA
Read-only
Inspect

Status check for a join / stitch render. "Is it ready / done yet?" / "check on it" / "where is my …" + placeholder_id or var2 link, media word = a join / stitch render → THIS tool. Status question with an id/link: NEVER answer null — one var2_get_*result always matches; pick by MEDIA WORD. Action verbs ("animate it", "upscale", "extend it", "make a 3d model from it") → the matching create/modify tool with that url/id, NOT a status tool. Pair: var2_join_videos. One immediate snapshot, never loop. NOT for routine progress checks: the inline preview in the chat updates itself when the job finishes — call this only when the user asks, or when you need the finished URL for a next step (then pass wait_seconds so ONE call waits for it - never check in a loop). Plain video renders → var2_get_video_result. A jp… id is a PLAN, not a render — it STARTS rendering via var2_join_videos({plan_id}); this tool takes only the plain-UUID placeholder_id of an already-queued render (from var2_join_videos or var2_render_timeline).

Inputs: placeholder_id (required), wait_seconds.

var2_confirm_uploadAInspect

Validate + register a file uploaded via var2_request_upload: probes the object, rejects/deletes bytes whose magic-byte signature or size does not match type, records it in the user's library → real placeholder_id for var2_join_videos chaining. Call with path + type from the request_upload response AFTER the PUT succeeds. Optional — skip if you only need public_url.

Inputs: path (required), type (required), name, bucket.

var2_create_3dAInspect

3D mesh (.glb) from ONE image. "Make a 3d model of this image/character" → THIS tool, even when the image is a www.var2.ai/image/ share link (action on an image routes here, not to a status tool). image: storage URL, var2 share link, or any https URL — pass the link the user gave. Models: trellis-2 (default, best price) | tripo ("high quality / detailed / production" 3D, PBR, ≤20k faces, pricier). No image at all → var2_create_image first. Local/attached file, no URL → var2_upload_asset first. Never fabricate the image URL.

Inputs: image (required), model, resolution, webhook_url, +1 more.

var2_create_audioAInspect

Create, COVER, EXTEND, or edit MUSIC / a song / a track (Suno). type ∈ create-music (default) | cover-music | extend-music | replace-music-section. "Cover this song in a " / "redo my track as jazz" / "same melody, new style / new lyrics" → type=cover-music with the song in audio_url (an https URL — upload a local file with var2_upload_asset first) OR audioRecordId (a var2 track); style = the new style; lyrics optional (empty keeps the original words); instrumental=true drops vocals. Source ≤ 8 minutes; returns 2 takes. Copyrighted recordings are NOT allowed as the source: Suno fingerprints it and rejects commercial songs ("matches an existing recording"). Only the user's own or royalty-free songs, e.g. a var2 track. "extend that track by N seconds" / "make the song longer" / "continue past 1:00" → type=extend-music, track id in audioRecordId. "That track" without a re-pasted id = the track from earlier in this conversation — use its placeholder_id from context; never answer null / ask for a link. "replace / swap seconds X-Y of the track" → type=replace-music-section. ANY make/extend/lengthen/replace-section on a song / track / music = THIS tool — never a video or image tool, never an invented tool. "A song AND cover art" → START HERE (art = follow-up var2_create_image, not an upload). customMode=false: prompt = song description ≤2800 chars on V6/V6_WILD (≤500 on V5_5 and older). customMode=true: prompt = full lyrics; style + title required. Never name real artists or songs (copyright filter).

Inputs: type, model, style, title, +13 more.

var2_create_dialogAInspect

TTS / voice-over (ElevenLabs v4 by default; model "elevenlabs-v3" for the previous generation, ≤5000 chars). dialogue = [{voice, text}, ...]. Default voices: EXAVITQu4vr4xnSDxMaL (Sarah, warm female), JBFqnCBsd6RMkjVDRZzb (George, warm British male). Hebrew: language_code "he" + nikud. Combined text ≤10000 chars. v4 reads English audio tags in square brackets inside text, before the words they color — [excited], [whispers], [sarcastic], [laughs], [sighs] — write them yourself, or pass auto_tag: true to have var2 add them. NOT for voice browsing — that's var2_get_voice_list.

Inputs: dialogue (required), model, title, auto_tag, +6 more.

var2_create_imageAInspect

Create a STILL IMAGE from a text prompt — "make/generate/draw an image of X", variations of an image — or edit one (type=image-to-image + image_url; image_refs[] locks character/style). A storyboard / N frames / N panels / comic strip with a consistent character = still IMAGES, one call per frame, not a video. Use var2_list_models for the complete current model list and supported settings; do not infer model availability from examples. Upscale/"sharper" → var2_modify_image; animate → var2_create_video; 3D → var2_create_3d.

Inputs: prompt (required), type, model, image_url, +7 more.

var2_create_motion_graphicsAInspect

Motion graphics with PIXEL-EXACT text: kinetic typography, animated titles, lower thirds, stat counters, logo stings, product / app launch videos, sale and quote cards, type over the user's own footage or photos. Drawn by code (0 % video generation), so words never garble and stay editable. USE THIS (not var2_create_video) when the deliverable is designed TEXT / graphics in motion; use var2_create_video for filmed-looking footage of people, places or scenes. READ var2_motion_catalog FIRST (styles, components, joins, physics). Modes: (1) composition — assemble reusable moments and motivated handoffs into a coherent film; free and deterministic. Search a varied shortlist, preserve coordinated mini-scenes, and inspect exact copy/timing in a free preview; (2) template + params (+ aspect_ratio, brand_kit_id) — preserve an entire requested template; free; (3) brief — the compiler writes the spec; put the exact on-screen words in quotes, one beat per line, plus mood / colors / pace (billed per second); (4) project_id (from any earlier motion result) or a spec you wrote, + instruction — revise it in plain language (billed like a compile); without an instruction, render:true renders it as-is (free). Media: image_urls (logos / photos → image layers, cutouts, frames, chips in the copy), footage_url (video or photo BEHIND the piece). Narration: voice_over {script|dialogue, voice} or voice_asset_id — every spoken phrase lands on screen as it is said; voice_convert re-voices it. Draft first: render:false (default) saves the piece as a draft (project_id, free) and returns warnings and human_summary (every word with its timing) while the user watches it play live in the chat; once they approve (the draft's "Render video" button posts "Looks good - render the final video (project )." - render THAT project), var2_save_motion_graphics { project_id, render: true } queues the MP4 and the chat shows it when done. Free alternatives: var2_preview_motion_graphics (compose or validate a spec) and var2_save_motion_graphics (editable project, optional free render). Template ids and params: var2_motion_catalog. Presets (briefs): kinetic, launch, stats, quote, list, countdown, lower-third, announcement.

Inputs: font, name, spec, brand, +19 more.

var2_create_videoAInspect

Create video. Replace/swap something in an existing clip, recast it with a new character, or restyle it into a preset look (anime, claymation…) → use var2_magic_replace instead. Pick type by INPUT: existing clip — www.var2.ai/video/ or video link (free-form prompt edit, "keep the motion") → video-to-video (video_url), NEVER image-to-video for a /video/ link. ONE image to animate → image-to-video. 2-3 image links, one consistent character in a NEW scene → reference-to-video (reference_image_urls), not kling_elements. Portrait image + audio track (talking head) → audio-to-video. Text only → text-to-video. "Turn this image into a video" / "animate this image" — link or URL included — is THIS tool, type=image-to-video, URL as first_frame_url (grok-imagine: reference_image_url); an action verb on an image link beats the /image/ path — NOT var2_get_image_result. Video only — songs/music are a different tool. Models: veo-3.1 (default), ltx-2.3, ltx-retake, wan-2.7, seedance-2, seedance-2.5, seedance-2-mini, happyhorse, kling, kling-3, kling-motion-control, kling-3-motion-control, gemini-omni-video, gemini-omni-flash-1-1, minimax-h3, wan-3, wan-3-prime, kling-3-omni, kling-4, sora-2, grok-imagine, pruna-avatar (talking head), heygen-avatar-iv (photorealistic talking head). kling-3: optional first/last frame (first_frame_url/last_frame_url or image_urls) + subject refs (kling_elements). kling-3-omni: Kling 3.0 Omni — text, frames, ≤7 references or one reference video, multi-shot, @elements and video edits (video_url) up to 4K. kling-4: Kling 4.0 — references (≤10 images, ≤5 videos, ≤3 voices, ≤7 @elements) or keyframes (first / middle keyframe_urls / last), 3–30s, 720p/1080p, always with audio.

Inputs: prompt (required), mode, type, draft, +33 more.

var2_estimate_costB
Read-only
Inspect

Token-cost estimate for one or more planned create_* calls. Use when the user asks "how much".

Inputs: items (required).

var2_extract_video_endpointsAInspect

Extract the ACTUAL first and last image frames from a completed owned video. Returns reusable image assets for shared-frame shot chaining. Paid per extracted frame through the normal approval/credit flow; never estimates the endpoint from FPS. Pass an owned video Share link, ID, or storage URL; import external videos first. A stable idempotency_key is required outside the assistant.

Inputs: video_url (required), name, projectId, idempotency_key.

var2_get_3d_resultA
Read-only
Inspect

Status check for a 3D MODEL / mesh / glb. "Is it ready / done yet?" / "check on it" / "where is my …" + placeholder_id or var2 link, media word = a 3D MODEL / mesh / glb → THIS tool. Status question with an id/link: NEVER answer null — one var2_get_*_result always matches; pick by MEDIA WORD. Action verbs ("animate it", "upscale", "extend it", "make a 3d model from it") → the matching create/modify tool with that url/id, NOT a status tool. Pair: var2_create_3d. One immediate snapshot, never loop. NOT for routine progress checks: the inline preview in the chat updates itself when the job finishes — call this only when the user asks, or when you need the finished URL for a next step (then pass wait_seconds so ONE call waits for it - never check in a loop). "Is the 3d model ready? placeholder " → THIS tool with that placeholder_id.

Inputs: placeholder_id (required), wait_seconds.

var2_get_audio_resultA
Read-only
Inspect

Status check for AUDIO: a song / music / track / tune. "Is it ready / done yet?" / "check on it" / "where is my …" + placeholder_id or var2 link, media word = AUDIO: a song / music / track / tune → THIS tool. Status question with an id/link: NEVER answer null — one var2_get_*_result always matches; pick by MEDIA WORD. Action verbs ("animate it", "upscale", "extend it", "make a 3d model from it") → the matching create/modify tool with that url/id, NOT a status tool. Pair: var2_create_audio. One immediate snapshot, never loop. NOT for routine progress checks: the inline preview in the chat updates itself when the job finishes — call this only when the user asks, or when you need the finished URL for a next step (then pass wait_seconds so ONE call waits for it - never check in a loop).

Inputs: placeholder_id (required), inline_media, wait_seconds.

var2_get_characterA
Read-only
Inspect

Fetch a saved character/product/style sheet by name ("use my character Dana", "the avatar we saved", "our hero product") — returns reference image urls to pass as image_refs / reference_image_urls for a consistent look. "Which characters do I have saved?" / "list my saved sheets" → call with NO name. Saved character referenced → call this FIRST, never regenerate from a text prompt.

Inputs: name.

var2_get_dialog_resultA
Read-only
Inspect

Status check for a VOICE-OVER / narration / dialog / spoken TTS clip. "Is it ready / done yet?" / "check on it" / "where is my …" + placeholder_id or var2 link, media word = a VOICE-OVER / narration / dialog / spoken TTS clip → THIS tool. Status question with an id/link: NEVER answer null — one var2_get_*_result always matches; pick by MEDIA WORD. Action verbs ("animate it", "upscale", "extend it", "make a 3d model from it") → the matching create/modify tool with that url/id, NOT a status tool. Pair: var2_create_dialog. One immediate snapshot, never loop. NOT for routine progress checks: the inline preview in the chat updates itself when the job finishes — call this only when the user asks, or when you need the finished URL for a next step (then pass wait_seconds so ONE call waits for it - never check in a loop). "Is the voice-over done?" → THIS tool, never var2_get_video_result or var2_get_audio_result.

Inputs: placeholder_id (required), inline_media, wait_seconds.

var2_get_image_resultA
Read-only
Inspect

Status check for an IMAGE (picture / photo). "Is it ready / done yet?" / "check on it" / "where is my …" + placeholder_id or var2 link, media word = an IMAGE (picture / photo) → THIS tool. Status question with an id/link: NEVER answer null — one var2_get_*_result always matches; pick by MEDIA WORD. Action verbs ("animate it", "upscale", "extend it", "make a 3d model from it") → the matching create/modify tool with that url/id, NOT a status tool. Pair: var2_create_image / var2_modify_image. One immediate snapshot, never loop. NOT for routine progress checks: the inline preview in the chat updates itself when the job finishes — call this only when the user asks, or when you need the finished URL for a next step (then pass wait_seconds so ONE call waits for it - never check in a loop).

Inputs: placeholder_id (required), inline_media, wait_seconds.

var2_get_resultsA
Read-only
Inspect

Show 2+ finished results in one gallery (browse / sort / group / filter). Triggers: "show me all of those", "put these three side by side", "line them up so I can compare", "drop / gather all of those clips into one place" — 2+ existing var2 links or ids = THIS tool, not an upload. items: 2–24 placeholder_ids and/or var2 links, any mix of images / videos / tracks / 3D. Finished items show instantly; rendering ones fill in. Never call with empty items. Single result → var2_get__result. "What models are available" has no ids → var2_list_models (voices: var2_get_voice_list). display=false returns data (states + URLs) without rendering, for chaining.

Inputs: items (required), title, display.

var2_get_video_resultA
Read-only
Inspect

Status check for a VIDEO clip (not a song / voice-over / stitched render). "Is it ready / done yet?" / "check on it" / "where is my …" + placeholder_id or var2 link, media word = a VIDEO clip (not a song / voice-over / stitched render) → THIS tool. Status question with an id/link: NEVER answer null — one var2_get_*_result always matches; pick by MEDIA WORD. Action verbs ("animate it", "upscale", "extend it", "make a 3d model from it") → the matching create/modify tool with that url/id, NOT a status tool. Pair: var2_create_video. One immediate snapshot, never loop. NOT for routine progress checks: the inline preview in the chat updates itself when the job finishes — call this only when the user asks, or when you need the finished URL for a next step (then pass wait_seconds so ONE call waits for it - never check in a loop). Join/stitch renders → var2_check_join_status; voice-overs → var2_get_dialog_result. "Is this ready yet? www.var2.ai/video/" → THIS tool.

Inputs: placeholder_id (required), wait_seconds.

var2_get_voice_listA
Read-only
Inspect

Voice catalog (ElevenLabs). "What voices do you have", "list voices", "which voice for <language/style>", "browse voices" → THIS, never null. source = account (~10-50 curated) | library (10K+ searchable). Filter: gender / language (ISO 639-1; "he" = Hebrew-verified) / category. Lists voices only — speaking is var2_create_dialog.

Inputs: age, live, accent, gender, +9 more.

var2_join_videosAInspect

Stitch videos / images / audio into one mp4 on a 30fps timeline. Fractional seconds are supported and rounded to frames, not whole seconds; the plan returns timeline_frames and fps. For a VERIFIED duplicated shared endpoint between 30fps clips, shorten the preceding clip by exactly 1/30 second, preserve the next first frame, and use no transition. A shared reference image does not prove actual endpoints match or eliminate duplicate frames; this tool does not extract or compare frames. Never replace one-frame removal with an arbitrary 0.2–0.5 second trim. Continuous background audio spans the joined timeline. Two-pass: dry_run=true (default) returns a plan to review with the user; dry_run=false renders it. Poll with var2_check_join_status (NOT var2_get_video_result).

Inputs: segments (required), resolution (required), aspect_ratio (required), name, +6 more.

var2_list_modelsA
Read-only
Inspect

Model catalog + pricing. "What/which models are available", "show me <image|video|audio|3d> models", "any other options", "browse models" → THIS, even when a modality word like "3d" appears. Filter by modality (image | video | audio | 3d | modify). Not a generator; var2_get_results is only for showing 2+ finished assets.

Inputs: modality.

var2_magic_replaceAInspect

Magic Replacer — edit an EXISTING video clip, keeping its motion. Engine: Seedance 2.5 (default) or MiniMax Hailuo 3 (engine: "minimax-h3"). Three modes: swap (Object Swap) replaces ONE named element — a product, person, outfit or prop — with what the reference image shows and keeps everything else (e.g. "put my serum bottle in her hand instead of the coffee cup", "replace the man in the red jacket with this character", "change her dress to this outfit"). transfer (Motion Transfer) keeps ONLY the motion, camera and timing and rebuilds cast + world from the references (e.g. "same dance, but this character on a beach"; keep_location=true keeps the original place). restyle keeps the people and motion and re-renders the look in a preset (e.g. "make this video anime / claymation / noir"). Use THIS tool whenever the input is an existing clip and the ask is replace / swap / recast / restyle. Use var2_create_video instead for NEW footage (text/image/reference-to-video), wan-2.7 for a free-form prompt edit of a clip, and ltx-retake to redo only a time range of a clip. Limits — seedance-2-5: source 4–30s, ≤8 reference images, 480p | 720p (default 720p), sound optional, billed per output second (= source seconds). minimax-h3: source 4–15s at 24–60 fps, ≤9 reference images (each beyond 5 costs extra), 768P | 2K (default 768P), always generates sound, billed per second of output + source clip. Trim longer clips first; output length and aspect ratio follow the source; upscale afterwards for more resolution. Do not pass a duration — the server measures the clip. Use the default engine unless the user asks for MiniMax / Hailuo / H3 or 2K output. Tips: character refs work best as a clean full-body photo or a multi-angle character sheet on a plain background (make one first with var2_create_image, model gpt-image-2, character-sheet prompt). In swap mode name the target concretely ("the blue mug on the desk") — or call first with analyze_only=true (free) to get the detected subjects with ids like "s2" plus a credit quote, then pass target:"s2". The scene analysis is saved per clip, so later calls on the same video_url reuse it instantly and the subject ids stay the same. For a cost estimate without analysis use var2_estimate_cost with model_id "magic-replacer", type "video-to-video", params {duration:, resolution, engine}. Returns a video placeholder (poll with var2_get_video_result). Reply in ONE short sentence + the share link; do not narrate tool calls.

Inputs: mode (required), video_url (required), engine, preset, +11 more.

var2_modify_imageAInspect

Remove background (cut out subject, transparent) OR upscale (2K/4K/8K, "sharper", "enhance", "isolate the ") an existing IMAGE — even when it is a www.var2.ai/image/ link. type ∈ upscale | remove-bg (no other type exists). IMAGE ONLY — judge by media type, not "edit"/"extend": track/song edits → var2_create_audio; video edits → var2_create_video / var2_join_videos. image_url: storage URL, var2 share link, or any https URL. Models: topaz-upscale (upscale_factor 2–6, default 2; credits scale with output megapixels), recraft-upscale (≤2048px), remove-background. Stylistic edits → var2_create_image type=image-to-image.

Inputs: image_url (required), type, model, webhook_url, +1 more.

var2_motion_buildAInspect

Paid (per second of motion + voice-over). Build the proposal the user approved in var2_motion_plan: narration, then the motion spec, saved as an editable piece (studio_url). Music / background video must already be urls (library, uploads, or made first with var2_create_audio / var2_create_video — the tool says when). render:true also queues the MP4.

Inputs: conversation (required), name, render, proposal, +2 more.

var2_motion_catalogA
Read-only
Inspect

READ FIRST for motion graphics (free). Choose section=styles, then section=components with query, style_id, tags, avoid_ids and recent_families for a compact ranked shortlist. component_id gives slots, physics, joins and constraints. Compose varied moments instead of repeating a whole template. The catalog includes every spec field with range + default, looks, effects, decor, media, finish, templates (params + a ready call), validated recipes, timing and pricing. Start with section=overview, then workflow; template_id = one template in full. available:false = not enabled for this user: do not offer motion.

Inputs: tags, limit, query, format, +7 more.

var2_motion_planA
Read-only
Inspect

Free. Plan a motion-graphics piece WITH the user before building: send their request (+ references); you get 2–5 follow-up questions (tone, audience, voice-over and who speaks, music, background video, style, length) to ask them, then a proposal (concept, beats with exact on-screen words, voice, music, footage, price). Pass conversation back unchanged with answers or a note. Build the approved proposal with var2_motion_build.

Inputs: note, prompt, answers, language, +5 more.

var2_preview_motion_graphicsA
Read-only
Inspect

Free, no side effects: compose catalog moments with COMPOSITION, build a TEMPLATE, or validate a hand-written SPEC exactly as the renderer sees it. For creative work prefer composition; read styles/components/joins first. Returns the normalized spec, warnings per layer, and every on-screen word with its timing (human_summary). Iterate until warnings is empty, then var2_save_motion_graphics.

Inputs: spec, params, template, composition, +2 more.

var2_read_motion_graphicsA
Read-only
Inspect

Free, owner-only. Read the current motion piece before editing. Start with view:"summary" for its layer IDs and revision, then layer_ids for up to four complete layers. Use typed layer_patches for small edits. The legacy default/full view returns the complete spec and can exceed model context; never save a truncated result.

Inputs: project_id (required), view, layer_ids.

var2_record_voiceA
Read-only
Inspect

Record the user's VOICE right in the chat (a recorder widget with Record / Stop / play back). "Let me record my voice", "record me", "I want to say it myself", "voice to voice from my recording", "change my voice" with no audio file → THIS first. The recording uploads to var2 and the widget sends its URL and length back to the chat; then call var2_voice_to_voice with it. Optionally pass the target voice (voice_id + voice_name from var2_get_voice_list) so it is carried along. Free; nothing is generated until voice-to-voice runs.

Inputs: voice_id, voice_name.

var2_render_timelineAInspect

Build a PRECISE multi-track video timeline — explicit per-clip start times, overlaps, per-clip mute/volume/fades/trim, TEXT/CAPTION overlays with styling, audio tracks with offsets, and MOTION-GRAPHICS pieces as transparent overlays on top of the video (lower thirds, titles, end cards: a "motion" track) — and save it as an editable project (render=false, default) or queue the render (render=true). Accepts a simplified tracks[] or the raw GUI editor schema; captions turns word/line lyrics timestamps into a styled caption track server-side. Frames at 30fps are canonical; seconds fields are converted exactly. For simple back-to-back concatenation use var2_join_videos instead; status of a queued render → var2_check_join_status.

Inputs: name, assets, render, schema, +4 more.

var2_request_uploadAInspect

Signed URL to upload a LOCAL file straight into var2 (durable, first-party, no expiry). Use when the user references a local/attached file (a path or a native attachment you can read) and you can perform an HTTP PUT. Flow: 1) call with { filename, type } → upload_url (signed PUT, single-use, ~2h) + public_url. 2) PUT the raw bytes to upload_url (curl -X PUT --data-binary @./logo.png -H "Content-Type: image/png" "<upload_url>") — bytes never pass through model context. 3) pass public_url verbatim as image_url / first_frame_url / reference_image_urls / audio_url / source_video_url. PUT blocked (sandboxed egress / host_not_allowed)? Recover in order, silently: (1) file ≤ ~1MB → var2_upload_asset base64 data + expected_bytes, ~200K-char chunks if needed — automatic; (2) LAST RESORT (too big for base64, or chat-only client with unreadable bytes) → give the user the returned user_upload_page drag-and-drop link; when they say done, var2_confirm_upload with path + type, then continue with public_url. Never downscale a file to dodge a blocked PUT. Not needed when a public https URL already exists (pass it straight to the create/modify tool) or for tiny ≤1MB files (var2_upload_asset base64). Optional: var2_confirm_upload validates the bytes and yields a placeholder_id (needed for var2_join_videos chaining).

Inputs: type (required), filename, content_type.

var2_save_characterAInspect

Save a CHARACTER SHEET (or product/style sheet) for reuse across sessions — a named set of reference images. Triggers: "save this character as Dana", "keep these angles of the avatar", "save the coffee cup as our hero product". Copies images into the in-app element library, returns durable urls; same name again replaces the sheet. UGC flow: hero image → upscale → image-to-image angles → save here. One-off uploads → var2_upload_asset.

Inputs: name (required), images (required), kind, notes.

var2_save_motion_graphicsAInspect

Free: save a COMPOSITION of catalog moments (+ brand_kit_id), a TEMPLATE (+ params), or a SPEC as an editable project; returns editor_url ("Refine in Motion Studio" opens it in the Studio). The user sees it play as a live draft in the chat. render:true also queues the MP4 (free; the chat shows the video when done) - after the user approved the draft. project_id updates / renders an existing piece in place. No compiler, no credits.

Inputs: name, spec, brand, params, +9 more.

var2_trim_audioAInspect

Cut or SPLIT AUDIO (a song / track / recording) by time — the only tool for that. Triggers: "cut seconds X to Y out of my song", "split this track into pieces", "keep only 0:12–0:40", "first 45 seconds", "shorten to 60s for pruna-avatar / heygen-avatar-iv". Modes: one range (start/end); explicit segments in ONE call (segments=[{start_seconds,end_seconds,id?},…]); SMART lyrics split (lyrics_timestamps + target/min/max_segment_seconds — never cuts mid-word, prefers line/chorus ends). Returns NEW asset(s) (url + placeholder_id); original untouched. audio_url: storage URL, var2 share link, or placeholder_id. Making new/longer music (extend, replace a section) → var2_create_audio; voice-overs → var2_create_dialog.

Inputs: audio_url (required), name, segments, end_seconds, +6 more.

var2_update_motion_graphicsAInspect

Free, owner-only. Prefer bounded layer_patches plus expected_updated_at from a fresh summary read: edits exact copy, solid colors and fonts while preserving all untouched data. Revision conflicts require re-reading. Legacy full spec replacement remains available only when you have the complete current spec. Never reconstruct a truncated spec.

Inputs: project_id (required), name, spec, layer_patches, +1 more.

var2_upload_assetAInspect

Upload a file the user ATTACHED to the chat (a photo from their phone, "my logo as a PNG", a voice memo) when NO link exists for it yet. If the message already contains a link or placeholder_id for the file, this tool is WRONG — a www.var2.ai / api.var2.ai / .supabase.co link is already inside var2; hand it to the matching create/modify/get tool. Sole pasted-link exception: NON-var2 link + explicit "import / pull it into var2". Text-only request, no file ("I want a song and cover art") → the matching create_ generator. When it IS an upload: call this FIRST, then pass the returned url verbatim as image_url / first_frame_url / reference_image_urls / audio_url / source_video_url. If your host filled in file (a native attachment object), just send it as-is — no url, no base64. Stores bytes durably in var2 storage; returns a first-party url (no expiry) + placeholder_id for var2_join_videos chaining. ≤100MB per file. base64 data: ALWAYS pass expected_bytes (+ sha256 if known) so truncation is rejected; >1.4M chars → parts via chunk_index/chunk_total/upload_id. Never pre-shrink a file to fit.

Inputs: type (required), url, data, file, +7 more.

var2_video_effectAInspect

Add a designed VIDEO EFFECT to a talking-head clip: graphics that sit INSIDE the footage - behind the person, in front of them, following their head - or that treat the footage as an object (a phone, an app window, a card). Use for "cards orbiting my head", "put a title behind me", "magazine cover", "viewfinder zoom", "pull back to show it in an app", "sticker outline", "checklist beside me". The first use of a clip prepares it once (the person is cut out frame by frame, in var2's own renderer); later effects on the same part of the clip reuse it. SEVERAL effects on one clip = ONE call with sequence: [{ effect, from_seconds, to_seconds, params }] - one piece, a different effect on each part of the clip (a call per effect makes a separate video per effect). The clip's length is read from the file: a step that runs past its end is refused with the real length. Returns an editable motion piece (editor_url) and a timeline_clip for var2_render_timeline; render=true also starts the finished video (the clip with the effect and the clip's sound): the answer names it in placeholder_ids and var2_check_join_status waits for it. If it answers state "preparing" (first use of a clip: it is being cut out, which takes a few minutes for a long clip), call it again with the same arguments: it picks up the same preparation and never starts a second one. Tell the user it is on its way; never loop. INSTANT AND FREE (graphics that only lie over the clip; nothing is prepared, saved at once): vfx-phone-card, vfx-checklist, vfx-name-tag, vfx-speech-bubble, vfx-arrow-callout, vfx-confetti, vfx-progress-bar, vfx-cta-button, vfx-letterbox, vfx-stamp, vfx-headline-top, vfx-check-pills, vfx-tier-list, vfx-film-frame, vfx-phone-fan, vfx-comment-reply, vfx-timeline. Every other effect prepares the clip first; when that has a price the answer says it (preparation_credits, charged once per clip), and an effect from the instant list never has one. Instant effects can be on screen TOGETHER (bars and a name tag): one call each without render, then one var2_render_timeline with a motion track per piece (overlay: true); a sequence only puts effects one after another. Texts on screen are SHORT - a label about 25 characters, a heading about 30, a row about 35, a tag about 14: a longer one is cut at a whole word and ends with "...", and the answer's warnings name each text that was cut. Effects:

  • vfx-orbit: Cards circle the speaker in real 3D: behind the body on the far side, in front on the near side. Each card is drawn by the engine from words, a picture, or both, and the cards can fly in and out on a spiral. Params: cards (up to 10 cards, three kinds: "a short title" (words only), "https://…" (picture only), or { title, subtitle, eyebrow, image, color } (words over / under a picture). Fewer than 6 are repeated to fill the ring); layout (ring (default: around the chest) | crown (a ring above the head) | arc (cards drift across an arc in front of the chest) | gaze (a ring across the frame above the speaker, at the place they look at most; it stays still, and its near cards pass in front of the hair)); enter (spiral (default: cards fly in one by one) | open (the ring grows from its centre)); exit (spiral (default) | open); facing (spin | tilt | camera (default spin)); card_layout (stack (picture above the words, default) | overlay (words over the picture)); text_size (size of the words on the cards, 0.5–2 (default 1). A card object may carry its own text_size); seconds_per_lap (default 4); radius (gaze layout only: half the width of the ring, percent of the frame width. Default: the whole width of the frame. Smaller (minimum 22 upright / 12 landscape) and it sits toward the side they look at most; it never moves during the clip).

  • vfx-screen-wall: A drifting wall of pictures fills the room behind the speaker. Params: images (2–40 image URLs).

  • vfx-ui-world: The whole frame becomes an app page: the clip plays in a rounded tile and the speaker's head rises out of its top edge; thumbnails drop into a prompt box and a command is typed. Params: greeting (line above the tile, e.g. "Afternoon, Simone"); command (the text typed into the box, e.g. "/videoeffects"); status (small status pill, e.g. "Running skill · video-effects").

  • vfx-frame-break: The room dims, the shot shrinks into a phone frame behind the speaker, and the speaker stays full size in front of it, breaking out of the frame. Params: device (phone | card (default phone)).

  • vfx-digital-zoom: A camera viewfinder (corner marks, battery, REC light, zoom readout) while the shot punches in on the face in steps. Params: zoom (strongest punch-in, 1.15–1.7 (default 1.45)).

  • vfx-phone-card: An app sheet in front of the speaker, rising from the bottom of the frame (beside the speaker, at the top, in landscape): steps tick off one by one, then the button turns green. Params: title (sheet heading); steps (up to 5 short steps; "text | tag" adds a right-hand tag); cta (button label).

  • vfx-checklist: A translucent dark panel in front of the speaker: a header with a live "2 of 5" count, then rows that tick green one by one, each with a small tag on the right. Params: title (panel heading); items (up to 6 rows; write "text | tag" to add a right-hand tag); position (bottom | top (default bottom)).

  • vfx-headline-wall: Newspaper clippings drop in around the head and pile up behind the speaker; the last one lands in front. Params: headlines (up to 8 short headlines).

  • vfx-zoom-through: A crash zoom into the speaker's face with motion blur and a flash: a cut into the next shot. Put it on the last half second of a clip. Params: strength (how far it pushes in, 2–5 (default 3.4)).

  • vfx-chart-grow: A glowing line chart draws itself behind the head; each point shows its value and the last one lands as a bold tag. Params: points (3–8 values (any scale)); color (line colour); prefix (before each value, e.g. "$"); suffix (after each value, e.g. "K", "%"); values (false hides the value labels).

  • vfx-pull-back: The camera pulls back from the shot to show it is a card on an app page (file name, Ready badge, clip chips, buttons), and the whole page tilts in 3D. Params: file_name (e.g. "reel_final.mp4"); badge (e.g. "Ready"); button (main button label, e.g. "Download").

  • vfx-green-screen: Split frame: the raw green-screen card on top, the cut-out speaker large below.

  • vfx-paper-cut: A scrapbook page: the speaker becomes a black-and-white sticker with a white edge on a yellow disc, caption words land as coloured pills, sticker shapes pop in around. Params: words (up to 8 caption words, shown as pills at the top); disc (disc colour).

  • vfx-doodle-boil: A wobbling hand-drawn outline around the speaker, with doodles popping beside the head. Params: color (line colour (default white)).

  • vfx-vertigo: A dolly zoom: the room pushes in and softens around the head while the speaker barely grows. Params: amount (0.04–0.14 (default 0.09)).

  • vfx-cover-story: A magazine cover: a giant title over the head with its foot tucked behind the hair, cover lines in front, a thin frame. Params: title (the masthead word); lines (up to 3 short cover lines); caption (one line under the speaker).

  • vfx-glass-type: A huge glass word fills the width behind the head and settles with a liquid wobble. Params: word (one word, up to 8 letters (a name works well)).

  • vfx-name-tag: A lower third: the name on an accent bar with the title under it, sliding in at the bottom left. Params: name (the name); title (role or handle (optional)).

  • vfx-spotlight: The room dims and softens and a pool of warm light sits behind the speaker, who stays sharp and bright. Params: strength (0.2–0.8 (default 0.5)).

  • vfx-halo: A glowing ring pulses behind the head and follows it. Params: color (ring colour).

  • vfx-caption-behind: Huge caption words land one at a time BEHIND the speaker, rising from behind the top of the head: the hair hides only the foot of each word, so it stays readable. Params: words (up to 12 words or very short phrases, shown one after another); frames_per_word (default: spread evenly over the clip); color (word colour (every second word uses the accent)).

  • vfx-speech-bubble: A comic speech bubble pops up beside the top of the head and moves with it; its tail grows out of the bubble and points down at the speaker. Params: text (what the bubble says (short)); side (left | right (default right)); color (bubble colour (default white)).

  • vfx-color-pop: The room turns black and white while the speaker stays in full colour.

  • vfx-background-swap: The room is replaced: the speaker stands in front of a picture you give, or a colour gradient. Params: image (https image URL for the new background (optional)); color (gradient start colour when no image is given); blur (0–20, softens the picture (default 4)).

  • vfx-echo: Ghost copies of the speaker, each a little larger, lag a few frames behind every move like a motion trail. Params: copies (1–3 ghosts (default 3)); lag_frames (frames between ghosts (default 3)).

  • vfx-neon-outline: A glowing neon line traces the speaker's silhouette while the room darkens. Params: color (line colour); width (4–16 (default 8)).

  • vfx-arrow-callout: A hand-drawn arrow and a label point at the speaker and move with them. Params: label (the label (short)); side (left | right (default left)).

  • vfx-big-stat: A huge number counts up behind the head (beside the speaker in landscape), with a label under it. Params: value (the number to count to); prefix (e.g. "$"); suffix (e.g. "%", "K", " tools" (6 characters at most)); label (one short line under the number (40 characters at most)).

  • vfx-confetti: Confetti bursts behind the speaker and a little falls in front: a celebration beat.

  • vfx-split-screen: The clip becomes a rounded tile on one side; a title and talking points land one by one on the other. Params: title (heading); points (up to 4 short points).

  • vfx-reaction-pip: A picture you give fills the frame and the speaker shrinks into a round bubble in the corner, reacting to it. Params: image (https image URL to show full screen (required)); corner (left | right (default left)).

  • vfx-sticker-burst: Your own pictures pop in around the head as white-edged stickers, some behind the speaker and some in front, and move with the head. Params: images (1–8 https image URLs (cut-outs with a transparent background look best)).

  • vfx-progress-bar: A thin bar across the top fills as the clip plays, so viewers see how much is left. Params: color (bar colour).

  • vfx-cta-button: A button pops up near the bottom, a cursor flies in and clicks it, and it turns green. Params: label (button text before the click, e.g. "Follow for more"); done (text after the click, e.g. "Following").

  • vfx-letterbox: Black cinema bars slide in from the top and bottom and a small title sits on the lower bar. Params: title (text on the lower bar (optional)).

  • vfx-countdown: Huge numbers count down behind the top of the head (beside the speaker in landscape), one per second. Params: from (start number, 2–9 (default 3)); color (number colour).

  • vfx-stamp: A rubber stamp slams onto the frame at an angle, in front of the speaker. Params: word (the stamped word, e.g. "NOPE" / "APPROVED"); color (ink colour (default red)).

  • vfx-scribble: The footage turns black and white, and a hand-drawn circle is scrawled round the face with a note beside it. Params: note (short handwritten note, e.g. "me →"); color (pen colour (default red)).

  • vfx-poster: A bold poster: black-and-white speaker, a giant red word behind, and three stacked lines in front. Params: word (the giant word behind); lines (up to 3 short lines in front); color (accent colour (default red)).

  • vfx-headline-top: A white band with a big headline; one word is highlighted. By default it lies over the bottom of the frame, in front of the speaker, and is added at once. position "top" puts the band above the shot and lowers the shot beneath it, so nothing of the speaker is covered: that prepares the clip like an effect behind the person (not instant). Params: headline (the headline (2–6 words)); highlight (one word of it to highlight); color (highlight colour); position (bottom | top (default bottom)).

  • vfx-check-pills: Blue pills with ticks pop up around the speaker one by one, like features being switched on. Params: labels (up to 6 short labels); color (pill colour).

  • vfx-tier-list: A tier list builds row by row in a dark panel in front of the speaker (S at the top, F at the bottom). Params: tiers (up to 6 rows; write "S | Claude" (letter | what goes there)).

  • vfx-film-frame: The clip sits inside a strip of film: black edges with sprocket holes and a stock name printed along the bottom. Params: stock (text printed on the film edge, e.g. "KODAK 400 · 21").

  • vfx-key-rain: Keyboard keys drop in around the head, some behind the speaker and some in front, and bob in place. Params: keys (up to 8 key labels, e.g. ["⌘", "Z", "K"]).

  • vfx-rabbit-hole: The clip appears inside itself again and again in white frames, each smaller: a tunnel of the same shot. Params: depth (2–3 nested frames (default 3)).

  • vfx-phone-fan: A hand of phones fans out across the bottom of the frame in front of the speaker, each showing one of your pictures. Params: images (2–5 https image URLs (screens or posters)).

  • vfx-face-lock: Camera focus marks snap onto the face and track it, with a small label under them. Params: label (small label under the marks (optional)); color (mark colour).

  • vfx-comment-reply: A viewer comment pops up above the head, then a reply slides in under it: the "comment X and I'll send it" moment. Params: comment (the viewer comment, e.g. "EFFECTS"); reply (the reply, e.g. "Sent you the pack"); name (commenter name (default "you")).

  • vfx-timeline: An editing timeline rises along the bottom: rows of clip blocks with a playhead sweeping across, and a title in the corner. Params: title (corner title, e.g. "AUTO-EDIT"); color (block colour (default red)).

Inputs: video (required), name, effect, params, +7 more.

var2_video_lookAInspect

See the person in a video: where their head is and WHERE THEY LOOK, second by second (camera, up, up-right, left…), plus the free room around them. Call it before placing graphics around a speaker ("put the cards where she is looking"). Prepares the clip once (same preparation var2_video_effect uses, charged like one export, reused afterwards). If it answers preparing, call again with the same arguments (it picks up the same preparation); never loop. Not a generator.

Inputs: video (required), to_seconds, aspect_ratio, from_seconds, +1 more.

var2_voice_to_voiceAInspect

VOICE TO VOICE (voice changer, ElevenLabs speech-to-speech): re-speak a recorded / uploaded voice in another voice, keeping its words, timing and emotion. Needs source_audio_url (a var2/https audio URL — from var2_record_voice, var2_upload_asset or a previous result), voice_id (var2_get_voice_list) and duration_seconds of the source (the recorder reports it). Up to 5 minutes. Priced per second of source audio. Returns a placeholder like other audio; the inline preview updates itself.

Inputs: voice_id (required), duration_seconds (required), source_audio_url (required), title, +2 more.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Servers

    • A
      license
      Not graded
      quality
      C
      maintenance
      Remote MCP server that lets AI assistants read tracked competitive intelligence — competitor news, reviews, website changes, social posts and ads — and act on it by managing projects, competitors, battle cards, reports and canvases. Every answer carries a source and a date, and write actions only publish when explicitly shared.
      MIT
    • A
      license
      A
      quality
      D
      maintenance
      Enables brand visibility monitoring across major AI platforms like ChatGPT, Claude, Gemini, and Perplexity. It allows users to track visibility scores, analyze competitor data, and receive actionable insights to improve AI-generated brand recommendations.
      16
      9 npm
      1
      MIT
    • A
      license
      Not graded
      quality
      C
      maintenance
      Browse IndustryLens's published competitive-intelligence reports and head-to-head competitor comparisons from any AI agent — real, source-backed data.
      MIT
    Try in Browser

    Glama MCP Gateway

    Add one secure layer between your agents and this server.

    Resources