Launchscene
Server Details
Real tech launch videos, scene by scene: search, see each scene, get timing to build your own.
The last two health checks of this connector failed. If this server requires authentication, missing or expired test credentials may be the cause. A test profile lets Glama authenticate for health checks and discover tools; it is separate from your personal connections.
If you are the author, claim ownership, then add or update a test profile under Admin → Test Profile.
- Status
- Unhealthy
- Uptime
- 0.0% over the last 24 hours
- OAuth
- Works in Glama
- Last Tested
- Transport
- Streamable HTTP · MCP 2025-11-25
- URL
- ChatGPT
- –
- Claude
- –
TDQS
Scored across 46 tools
The set has several overlapping clusters where selection is genuinely hard: search_scenes vs search_by_moment vs pattern_search vs find_similar, get_scene vs get_scenes vs get_video vs get_video_structure, and list_formulas vs get_formula vs typical_structure vs plan_structure vs plan_from_references. The verbose descriptions do draw distinctions ("Main scene search", "One formula in full", order-preserving plan alignment), so an agent can usually recover, but boundaries require careful reading rather than being self-evident.
All names are snake_case with a predictable vocabulary (get_/list_/search_/plan_/save_/critique_/compare_ plus clear nouns). A meaningful minority are bare noun phrases (beat_grid, storyboard, trends, corpus_stats, music_brief, generation_prompts, readability_check), which breaks the verb_noun pattern but stays readable and unambiguous in style.
46 tools is heavy for what is essentially a video-corpus search plus planning/critique workflow, and it pushes well past the 25+ threshold. Several near-duplicate read tools (get_scene/get_scenes, get_video/get_video_structure, get_brand/get_brand_kit, list_formulas/get_formula) inflate the surface without adding distinct capability.
Coverage is impressively broad: discovery, retrieval, planning, storyboarding, render upload/measurement, brand kits, boards, exports and corrections all exist with lifecycle hooks (create_render_upload/cancel_render/get_render_result, list_renders/save_render). Gaps are minor but real — no create_board or remove_from_board, and no delete/rename for saved render projects, leaving the board and project side one-directional.
Publisher details
- Operator
- YC Strategic Ventures, publisher of Launchscene · Publisher source
- Operator website
- https://launchscene.app · Publisher source
- Vendor relationship
- First-party · Publisher source
- Documentation
- https://launchscene.app/mcp · Publisher source
- Privacy policy
- Not provided
- Trust center
- Not available
- Restrictions
- Requires a Launchscene account (sign-in by one-time email link). Free for every account during the beta; later part of Launchscene Pro. · Publisher source
Available Tools
46 toolsadapt_sceneAdapt a scene to your copyCRead-onlyIdempotentInspect
Substitute all source copy with your text, recalculate cadence and reflow normalized text layout. Every adapted value is extrapolated; source mid-presence boxes are not trajectories.
Inputs: texts (required), scene_id (required), aspect, duration_ms. Returns: schema_version
beat_gridBeat gridBRead-onlyIdempotentInspect
Calculate a synthetic constant-tempo beat/downbeat grid and snap proposed cut times while preserving exact start/end. Supply phase explicitly; BPM alone cannot recover an audio file’s real timing.
Inputs: bpm (required), duration_ms (required), cuts_ms, phase_ms, +3 more. Returns: schema_version
cancel_renderCancel a render analysisADestructiveInspect
Cancel your private render analysis and remove its uploaded input. A running worker stops its current subprocess and discards partial measurements. This cannot modify another account’s task or any catalogue video.
Inputs: task_id (required). Returns: schema_version
compare_groupsCompare two groupsCRead-onlyIdempotentInspect
Compare two explicit groups using video-weighted distributions. Reports overlap; no causal claims, quality scores or silent broadening.
Inputs: group_a (required), group_b (required), role, metrics, +1 more. Returns: schema_version
compare_to_referenceCompare a plan to a referenceBRead-onlyIdempotentInspect
Compare a plan to one chosen reference using order-preserving narrative-role alignment. Insertions and deletions stay explicit rather than forcing unrelated scenes to match.
Inputs: plan (required), video (required). Returns: schema_version
corpus_statsCorpus statisticsCRead-onlyIdempotentInspect
Describe a metric with dispersion, a minimum of eight distinct videos and cited median/extreme examples. Each video receives equal weight; missing measurements stay unknown.
Inputs: role, metric, filters, min_videos, +1 more. Returns: schema_version
create_render_uploadCreate a render uploadADestructiveInspect
Create a one-use account-private upload for a video you are authorized to analyse. Stream raw MP4/WebM bytes to the returned authenticated PUT endpoint; 100 MiB and 180 seconds maximum. This reserves temporary private storage and never publishes to the catalogue.
Inputs: no named parameters. Returns: schema_version
critique_planReview a planBRead-onlyIdempotentInspect
Contextualize storyboard durations and pacing against video-weighted corpus distributions. Asks questions, never assigns severity or a quality score. Sparse comparisons remain unknown.
Inputs: plan (required), filters, min_videos, allow_broaden. Returns: schema_version
critique_renderAnalyze an uploaded renderADestructiveInspect
Queue measurements for your successfully uploaded render: cuts, OCR, motion, audio and synchronization. Returns immediately with a task ID. No local path or third-party video URL is accepted. Requires a running worker; poll get_render_result for real measured results.
Inputs: task_id (required). Returns: schema_version
export_recipeExport starter codeARead-onlyIdempotentInspect
Generate editable starter source from an adapted recipe, or a reference plus mandatory replacement text: Remotion TSX, HyperFrames HTML, CSS, GSAP, lottie-web JSON or After Effects JSX importer. Optional style sets the background, text colour and font family (your own brand, or values read from get_brand_kit for study); the response reports their contrast ratio. No source footage/logos; unknowns stay adjustable. Generated code is not automatically render-tested.
Inputs: target (required), style, texts, aspect, +3 more. Returns: schema_version
find_similarFind similar scenesCRead-onlyIdempotentInspect
Scenes that look like a chosen one, from precomputed visual similarity, filtered to the published catalogue. Each of the first 12 results comes with a signed image.
Inputs: id, limit, cursor, images, +2 more. Returns: schema_version
generation_promptsShot briefsBRead-onlyIdempotentInspect
Turn a supplied structure into vendor-neutral shot briefs, keeping user overlay text separate from generated imagery and preserving reference links when supplied.
Inputs: plan (required). Returns: schema_version
get_audio_mapGet audio mapBRead-onlyIdempotentInspect
Read chronological beat/downbeat/accent/sound events, sections and bounded loudness/stem samples in a window. Events are paginated; numerical sound levels are measurements, not music files.
Inputs: slug, limit, cursor, end_ms, +3 more. Returns: schema_version
get_brandGet a brandBRead-onlyIdempotentInspect
Read a brand’s paginated launch references and descriptive style counts with distinct-video counts.
Inputs: slug (required), limit, cursor. Returns: schema_version
get_brand_kitGet a brand’s house styleARead-onlyIdempotentInspect
A brand’s house style as rules an agent can follow, counted film by film over its published launches: palette, type, scene formats, techniques, transitions, camera, opening, call to action, voice, length and pace. Each rule gives n of N films, how common the trait is across the whole library, its method and example films. Use it to plan a film in that brand’s style; it describes the catalogue, not the brand’s official guidelines.
Inputs: slug (required). Returns: schema_version
get_camera_trackGet camera trackBRead-onlyIdempotentInspect
Read measured camera movement intervals. Zoom uses scale ratios; pan/direction use fractions of image width. Easing is a heuristic class. Clipping does not remeasure amplitude.
Inputs: slug, limit, cursor, end_ms, +3 more. Returns: schema_version
get_formulaGet a formulaARead-onlyIdempotentInspect
One formula in full: each beat with its median share and duration at the median length, the techniques and scene formats at least a quarter of its films use in that beat, voice and pace, example films and a template prompt. Then plan your film with plan_from_references using its examples.
Inputs: id (required). Returns: schema_version
get_framesGet reference framesBRead-onlyIdempotentInspect
Get up to twelve reference frame URLs, or four bounded inline images. Array times are scene-relative. Requested and actual image times are always separate. Text overlays require precomputed images at the true OCR observation timestamp.
Inputs: at, id, limit, inline, +2 more. Returns: schema_version
get_historyGet MCP historyARead-onlyIdempotentInspect
Read this account’s recent MCP activity: operations and consulted scene IDs, without private briefs or credentials.
Inputs: limit, cursor. Returns: schema_version
get_render_resultGet render analysisARead-onlyIdempotentInspect
Read the progress or measured result of your private render task. Finished measurements are contextualized against published reference distributions (minimum eight videos); no quality score or invented observations. Source uploads expire after 24 hours.
Inputs: task_id (required), filters. Returns: schema_version
get_sceneGet a sceneARead-onlyIdempotentInspect
Read a card, annotated sheet or measured recipe. Recipes include relative and absolute timing, OCR box meaning, heuristic motion measurements, source provenance, verification and missing-data reasons.
Inputs: id, depth, scene_id. Returns: schema_version
get_scenesCompare scenesARead-onlyIdempotentInspect
Read up to six scene references together for comparison. Recipe batches are limited to three. Unavailable/private scenes return the same generic error.
Inputs: ids (required), depth. Returns: schema_version
get_text_trackGet on-screen text trackARead-onlyIdempotentInspect
Read paginated OCR text intervals clipped to a video window or a scene. Boxes remain attached to their original observation; old exports only permit an estimated box timestamp.
Inputs: slug, limit, cursor, end_ms, +3 more. Returns: schema_version
get_transcriptGet voice transcriptCRead-onlyIdempotentInspect
Read timestamped voice transcript segments in a video window or scene. Segment clipping does not imply word-level alignment.
Inputs: slug, limit, cursor, end_ms, +3 more. Returns: schema_version
get_videoGet a launch videoARead-onlyIdempotentInspect
Read a published video summary, style measurements and paginated scene cards. Video/media/music/font/logo files are never exported.
Inputs: slug (required), limit, cursor. Returns: schema_version
get_video_structureGet a video structureCRead-onlyIdempotentInspect
Read narrative sequences, durations, audio arc, voice, CTA and whole-video synchronization. A missing precomputed strip is explicitly reported.
Inputs: slug (required), limit, cursor. Returns: schema_version
list_boardsList boardsCRead-onlyIdempotentInspect
List this account’s private boards and a bounded set of currently published scene references.
Inputs: limit, cursor, board_id, items_limit. Returns: schema_version
list_brandsList brandsARead-onlyIdempotentInspect
List brands with published launch films. Includes no logos or brand media files. A brand that only has clips is not listed; its clips are found through search_scenes and get_brand.
Inputs: limit, query, cursor, industry. Returns: schema_version
list_formulasList launch video formulasARead-onlyIdempotentInspect
Structures shared by at least 8 published films each: the same order of beats (consecutive feature and demo passages count as one) in the same length band. Each comes with its film count, median length and example films. Optional filters: band and a beat it must contain.
Inputs: band, contains. Returns: schema_version
list_rendersList your private rendersARead-onlyIdempotentInspect
List the user’s private render projects: title, number of versions, latest video and the address of each project on Launchscene. Use a project id with save_render to add the next version.
Inputs: no named parameters. Returns: schema_version
music_briefMusic briefARead-onlyIdempotentInspect
Build a musical brief from a plan, with corpus BPM/voice-bed distributions and a timed proposed energy arc. Clearly separates proposed impacts from measured references; no audio is generated.
Inputs: plan (required), filters, min_videos, allow_broaden. Returns: schema_version
pattern_searchSearch scene patternsARead-onlyIdempotentInspect
Find consecutive scene patterns, optionally aligned with measured beats. Pattern steps use taxonomy keys, not guessed natural-language semantics. Returns individual cited matches; no small-sample trend claims.
Inputs: pattern (required), limit, cursor, repeat, +3 more. Returns: schema_version
plan_from_referencesPlan a film from referencesARead-onlyIdempotentInspect
Plan one film from 1 to 5 chosen reference videos and your brief: one reference is the spine (the one sharing most narrative roles with the others, or the one you name), its passages are scaled to your exact duration and its scenes become shots; the others are aligned role by role and offered as alternatives. Your copy goes to the matching roles (message to the hook, features to feature reveals in order, cta to the call to action). Returns a beat sheet and a plan that critique_plan, beat_grid, music_brief and generation_prompts accept as is.
Inputs: references (required), duration_ms (required), brief, spine, +2 more. Returns: schema_version
plan_structurePlan a film structureBRead-onlyIdempotentInspect
Create two or three contrasting exact-duration plans from cited observed sequences. Scaled timing is explicitly a proposal; under-sampled patterns are not presented as norms.
Inputs: duration_ms (required), aspect, filters, variants, +2 more. Returns: schema_version
readability_checkCheck text readabilityBRead-onlyIdempotentInspect
Check supplied text geometry and display time; compare size, reading demand and margins to per-video OCR distributions. Optional color contrast follows WCAG’s formula, with applicability caveats.
Inputs: items (required), filters, min_videos, allow_broaden. Returns: schema_version
reference_briefFrame-numbered reference briefARead-onlyIdempotentInspect
Turns one chosen scene, or a window of up to 120 s of a video, into a frame-numbered reference for motion work (REFERENCE.md in markdown): scene boundaries, text entrance, landing and readable hold, camera moves and audio cues, at the source frame rate and at your composition frame rate. Measured, annotated, derived and unknown values stay separate, and frames worth checking are listed.
Inputs: end_ms, scene_id, start_ms, target_fps, +1 more. Returns: schema_version
save_renderSave a render to your private spaceADestructiveInspect
Save the video you just rendered to the user’s private Launchscene space, with the brief and the Launchscene references it was built from. Call it once the file exists, and again with project_id for each later version of the same film. Returns a one-use upload command to run once with the real file path (MP4 or WebM, 100 MiB and 180 seconds maximum); the file is then kept in the project and measured against its reference on its own. Private to the account: never published, never added to the library. Give the user the returned project.url.
Inputs: note, agent, brief, title, +3 more. Returns: schema_version
save_to_boardSave to a boardBDestructiveInspect
Save published scenes to a private account board shared with the Launchscene website. Requires an account and write scope.
Inputs: scene_ids (required), note, board_id, board_name. Returns: schema_version
search_by_momentSearch timeline momentsCRead-onlyIdempotentInspect
Search measured/annotated timeline moments and test their offset from real beats or downbeats. Unknown alignment never matches an on-beat filter.
Inputs: label, limit, cursor, filters, +5 more. Returns: schema_version
search_scenesSearch launch scenesARead-onlyIdempotentInspect
Main scene search. Finds launch-video scenes from a plain description ("a cursor clicks Export and the file appears") and/or taxonomy filters. Each of the first 12 results comes with a signed image you can see (key frame and its progression), plus image_url and launchscene_url. Each result also carries its film (title, duration, scene count, order of passages) and the scene's position in it, so open get_video only for the full scene list. Only published references, each with its annotation_status; filters apply to each scene and unknown measurements are excluded. Scenes of clips are included (film.kind is "clip": a short loop or site footage that is a scene reference, not a launch film); filters.kind narrows to "film" or "clip".
Inputs: sort, limit, query, cursor, +2 more. Returns: schema_version
search_videosSearch launch videosARead-onlyIdempotentInspect
Main search for whole launch films. Finds published launch videos by description, brand, industry, timing, sound or style. Each of the first 12 results comes with a signed image of its scenes, colored by narrative role. Score is relevance, not quality. Clips (short loops and site footage that announce nothing by themselves) are left out unless filters.kind is "clip" or "any".
Inputs: sort, limit, query, cursor, +2 more. Returns: schema_version
storyboardStoryboard a planARead-onlyIdempotentInspect
Lay a plan out as a storyboard before anything is built: one panel per shot with its number, role, timing, your on-screen text, camera and transition, and the key frame of its reference scene. Returns the panels as data and, when images are on, one storyboard sheet image (the first 16 shots). Frames are references for framing and rhythm, never generated or the final render. Takes the plan from plan_from_references or plan_structure as is.
Inputs: plan (required), title, images. Returns: schema_version
suggest_correctionSuggest a correctionBDestructiveInspect
Submit an annotation correction to the existing human review inbox. Does not edit or republish the video.
Inputs: message (required), scene_id (required). Returns: schema_version
transitions_afterWhat follows a roleBRead-onlyIdempotentInspect
Describe what follows a narrative role: next role, transition, or techniques. Frequencies are distributions of proportions within each video, preventing long edits from dominating.
Inputs: after_role (required), kind, limit, cursor, +3 more. Returns: schema_version
trendsYearly trendsCRead-onlyIdempotentInspect
Year-by-year descriptive distributions. Each year independently requires enough distinct videos; sparse years remain null. Release dates are the catalogue published_at dates.
Inputs: role, limit, cursor, metric, +2 more. Returns: schema_version
typical_structureTypical structuresCRead-onlyIdempotentInspect
Observed narrative role sequences and per-video duration distributions, with alternative examples and sample-size suppression. duration_s sets a ±50% duration window unless explicit bounds are supplied.
Inputs: limit, filters, duration_s, min_videos, +1 more. Returns: schema_version
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Related MCP Connectors
Curated real launch videos split into timed sections (hook, problem, feature, proof, CTA).
Pick a real launch video as a template, get its frame-by-frame breakdown and plan your own remake.
4,100 launch videos measured shot by shot. Free account to search; Pro for blueprints and plans.
Create and edit product videos from a brief: launch, demo and feature films, rendered for you.
Related MCP Servers
- FlicenseNot gradedqualityCmaintenanceEnables searching and analyzing curated launch videos and timed sections, retrieving hooks and benchmarks, and creating and managing stitches and comments for planning and scripting videos.-

AI Video Atlasofficial
AlicenseNot gradedqualityCmaintenanceEnables AI agents to plan multi-shot AI video productions by splitting a plain description into beats and selecting a sourced shot, model, prompt, and estimated cost for each. Also lets clients search a library of real creator prompts, recommend models with evidence, and look up vendor-stated $/second pricing and prompting guides.MIT- AlicenseAqualityCmaintenanceEnables screenwriters to search for similar scenes, fetch scene details, list indexed movies, find mood reference scenes, and get analysis style guidance, all grounded in real produced scripts.5MIT
- AlicenseNot gradedqualityDmaintenanceEnables AI agents to produce professional promo videos by providing tools for capture, media search, scene writing, rendering, editing, and incremental revision.12 npm17MIT
Glama MCP Gateway
Add one secure layer between your agents and this server.