zoom-editor
Processes Zoom meeting recordings and their accompanying audio_transcript_*.vtt transcripts to produce short captioned videos, including clips, trailers, longform best-ofs, and social-sized versions with burned-in captions, waveform, spectrum bars, and headlines. Agents can create projects from Zoom recordings, inspect transcript cues, select moments, render videos, and retrieve completed job files.
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
zoom-editor
Turn a meeting recording and its transcript into short, captioned videos ready for social media: a single clip, a trailer of about 30 seconds, or a longform best-of. Each comes as a plain cut and, optionally, as 1:1, 9:16 and 16:9 versions with burned-in captions, a waveform, spectrum bars and a headline.
It runs as one service on port 8093, serving an MCP server (streamable HTTP at /mcp) for AI agents and a REST API (/api/v1) for scripts. It encodes on the GPU with NVIDIA's encoder (h264_nvenc) when one is available and falls back to the CPU (libx264) otherwise.
Zoom is the motivating case because Zoom cloud recordings come with an audio_transcript_*.vtt. Any video with a WebVTT or SRT transcript works.
How an agent uses it
create_project: point it at the video (URL, or a path under the media root) and the transcript.get_transcript: read the cues and decide which moments are worth sharing. The service doesn't pick highlights itself; the calling agent (or person) does, so no LLM key is needed here.Render:
render_clip: one moment.render_trailer: one whole sentence from each moment, the one that best matches itshook, played in recording order.render_longform: every moment in full, with the headline changing for each.render_social: no cut; the social sizes of the whole video, for a clip you already have.caption_startsays where it sits on the transcript's timeline, andheadline_timelinecan change the headline over time.
get_jobuntil the job'sstatusisdone, then download the file URLs.
Times can be given in seconds or as HH:MM:SS.mmm, measured on the source video's timeline. Captions never cut a sentence mid-way, and trailer snippets always start and end on sentence boundaries.
Related MCP server: levea-mcp-server
Run it
docker pull ghcr.io/isleprince/zoom-editor # or build: docker compose up -d --build
docker compose up -d # GPU passthrough requested; works without one too
curl http://localhost:8093/api/v1/health/api/v1/health reports which encoder was chosen and why. The choice comes from a short test encode at startup, not from checking nvidia-smi. For NVENC inside Docker, the NVIDIA container runtime must pass the video capability, which the compose file requests.
Connect an MCP client to http://<host>:8093/mcp (streamable HTTP).
Variable | Default | |
| 8093 | |
|
| projects and job outputs |
| (empty, off) | local files callers may reference; nothing outside it |
|
|
|
| (empty) | if set, required as |
| request host | base URL for file links in MCP results |
| 1 | concurrent jobs (one GPU encodes one job best) |
| 3 with NVENC, else 1 | social sizes rendered at once within a job (the CPU filter graph is the bottleneck on long videos) |
| 14 | finished jobs are deleted after this; projects stay |
| 8 | cap on downloads from |
REST
GET /api/v1/health
POST /api/v1/projects {video_url|video_path, transcript_url|transcript_path|transcript_text, name}
POST /api/v1/projects/upload multipart: video, transcript?, name?
GET /api/v1/projects[/{id}] DELETE /api/v1/projects/{id}
GET /api/v1/projects/{id}/transcript ?start=&end=
POST /api/v1/jobs {project_id, type: clip|trailer|longform|social, moments:[{start,end,label,hook?}],
sizes:["1x1","9x16","16x9"], subtitles, headline, accent, bg, quality,
caption_start, headline_timeline:[{t0,t1,text}]} (last two: social)
(a clip may instead give start/end/headline at the top level)
GET /api/v1/jobs[/{id}] GET /api/v1/jobs/{id}/files/{name}Develop
pip install -e '.[test]'
pytest -q # generates its own test video; needs an ffmpeg that has drawtextThe layout and transcript rules were ported from the nokemo.com Zoom-to-social pipeline (render_social.py, render_compilations.py) and behave the same.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
This server cannot be deployed
Maintenance
Related MCP Connectors
Clip videos into captioned shorts, add captions, and schedule posts from AI agents.
The agentic video editor for ChatGPT and Claude. Ask in your chat and it cuts, captions, adds music, titles and motion graphics on a real timeline you can open and adjust.
Video tools for agents (trim, captions, 9:16, GIF/MP3) + hire us for fixed-price jobs. No API key.
Turn long videos into captioned, reframed short clips, thumbnails, and social posts.
Related MCP Servers
- AlicenseAqualityDmaintenanceEnables AI agents to edit video assemblies from A-roll and B-roll, add captions, and publish to social media platforms.272,053 npmMIT
- AlicenseBqualityBmaintenanceEnables AI agents to turn natural-language creative direction, transcripts, and source media into fully structured, editable video projects, then verify and render delivery files.255MIT
- AlicenseNot gradedqualityBmaintenanceEnables AI agents to autonomously edit raw video footage into publish-ready videos with millisecond-accurate cuts, Whisper transcription, karaoke captions, motion graphics, and high-CTR thumbnails.7 npm6MIT
- FlicenseNot gradedqualityBmaintenanceEnables users to turn natural-language prompts and raw video clips into finished, beat-synced montage cuts with shot selection, music matching, subtitles, transitions, and rendering through compatible AI agents.-