Quick answer
Same category, different unit of work. Both are official MCP servers and both leave a real project you can open. invideo gives the agent a multitrack timeline; Descript gives it a transcript
Pick invideo if: you are assembling from footage and generated assets, need colour and audio work, or need a rendered MP4
Pick Descript if: your source is talking-head, podcast or screen recording, and the edit is really a text edit
Cost shape: invideo meters generation only - editing is free on every plan. Descript meters imports and AI actions
Free entry: Descript has a free plan, invideo does not. invideo starts at $9 a seat
The catch each way: Descript MCP cannot export an MP4 on request - its own docs call that “the big one”. invideo Basic cannot touch Seedance 2.5 at all
At a glance
| invideo Remote Agents | Descript MCP |
|---|
| What the agent edits | Multitrack NLE timeline | The transcript - text edits become cuts |
| Official server | Yes, mcp.invideo.io/mcp, Streamable HTTP | Yes, connector plus a published server URL |
| API key | None - account sign-in | None - account sign-in |
| Watch it work | Each edit lands live under a named cursor | Edits land in a project in your drive |
| Take over mid-edit | Yes, at any frame, alongside the agent | Yes, open the project and fine-tune |
| Generation built in | Images, video and audio | AI media and avatars, but the Claude directory connector excludes generative image and video - needs a custom MCP setup |
| Rendered MP4 from the agent | Yes | No - publish to a link with a download URL instead |
| NLE handoff | Premiere Pro, Final Cut Pro, DaVinci Resolve | EDL, AAF, SESX, FCPXML, Premiere XML, DaVinci Resolve XML |
| Free plan | None | Yes - $0, 60 min media a month, 720p, watermarked |
| Paid entry | $9 per seat a month | $16 per seat a month |
The real difference: what the agent is holding
Both tools answer the same instruction - “cut this down and clean it up” - but they hand the agent completely different objects, and that decides which one suits your footage.
invideo gives it a timeline. The agent works in tracks: it makes cuts, drops titles and transitions, runs colour correction, mixes audio, and generates images, video or audio to fill gaps. Six starter prompts ship with it - Rough cut, Color, Captions, Audio, Generate and Deliver - which map neatly onto an edit bay workflow. Every change appears live under a cursor labelled with the agent’s name, undo works per edit, and there are checkpoints to roll back to.
Descript gives it a transcript. Editing the words edits the video, so the agent’s natural moves are Remove Filler Words, Remove Retakes, Studio Sound, captions, clips, overlays, reframing and translation. It can import from a file URL or upload, read projects back, manage compositions, and track or cancel jobs. For an hour of someone talking, that is a far more direct route than asking an agent to find the right frame on a timeline.
The decision rule: if a human editor would start by reading the transcript, use Descript. If they would start by laying clips on a timeline, use invideo. Footage with no speech - b-roll, product shots, generated material - is invideo’s territory by default, because there is no transcript to edit.
How each one charges you
This is the least obvious difference and the one most likely to surprise you at the end of a month.
invideo meters generation only. The full timeline, multiplayer editing, colour, audio and 3D are on every plan including Basic, and invideo’s own plan calculator assigns no credit cost to 40 hours of timeline editing. An agent that makes two hundred cuts and three generations costs you three generations.
Descript meters both ends. Its own documentation is explicit: “Using the API or MCP can use your media minutes and AI credits.” Imports consume media minutes, and Underlord actions such as Studio Sound, Remove Filler Words or captions consume AI credits. So a long import costs you before any editing happens, and an agent told to clean up a two-hour recording spends on both axes.
Neither is better in the abstract. If your work is long recordings with light processing, Descript’s media hours are the binding constraint. If it is short pieces with heavy generation, invideo’s credits are.
Pricing, side by side
| invideo | Per seat / mo | What you get |
|---|
| Basic | $9 ($4 annual, current promo) | 190 credits. Full timeline. No Seedance 2.5, Agent Two Lite only |
| Pro | $30 ($15 annual) | 1,000 credits - about 10 five-second 1080p clips. Seedance 2.5, all agent tiers |
| Ultra | $80 ($40 annual) | 3,000 credits - about 31 clips. Lowest cost per credit |
| Descript | Per seat / mo | What you get |
|---|
| Free | $0 | 60 min media a month, 100 AI credits one-time, 720p, watermarked, no timeline export. No card needed |
| Hobbyist | $16 (or $24) | 10 media hours, 400 credits a month, 1080p, watermark-free |
| Creator | $24 (or $35) | 30 media hours, 800 credits, 4K, 1-3 seats |
| Business | $50 (or $65) | 40 media hours, 1,500 credits, 4K, up to 5 seats |
Descript is the only one of the two you can try without paying - but note what the free tier withholds. Timeline export is not included on Free, so the EDL and FCPXML handoff that makes Descript useful in a professional chain needs a paid plan, and local exports are capped at 720p with a watermark. invideo has no free plan at all, so its $9 Basic is the real floor for either tool if you want unwatermarked output.
Limits each vendor documents itself
Both publish their own constraints, which is more than most tools in this category manage. These are quoted from their documentation rather than inferred.
| invideo Remote Agents | Descript MCP |
|---|
| Credits do not roll over month to month | No rendered download without publishing - the docs call MP4 export “the big one” that is missing |
| invideo reserves the right to change credit costs without notice | Scoped to a single Drive - switching means logging out of Descript on the web and reconnecting |
| Basic has no Seedance 2.5 access and caps at Agent Two Lite | YouTube URLs are not supported as import sources |
| No API key means no way to cap an agent’s spend separately from your seat | Job history kept for 30 days only; transcript export has no DOCX |
| Requires a client that supports remote MCP servers | The Claude directory connector excludes generative image and video tools |
The two that will actually bite: Descript’s missing MP4 export, if your pipeline ends in a file rather than a link, and invideo’s lack of a per-agent spending cap, if you are letting an agent iterate unattended. Checkpoints undo the timeline, not the spend.
Which to choose
Take invideo if your material is footage and generated assets rather than speech, you need colour and audio work on a real timeline, you want the agent to generate as well as assemble, or the deliverable is a rendered MP4. The free-editing model also suits agents that iterate a lot on structure and generate rarely.
Take Descript if your material is people talking - interviews, podcasts, webinars, screen recordings - and the edit is removing filler, tightening retakes and fixing audio. Transcript-driven editing is simply the shorter path there, and it is the only one of the two with a genuine free tier to test the workflow on.
Take both if you produce a weekly show and marketing cuts from it. Descript for the episode, invideo for the promos. They cost $9 and $16 at entry, which is less than one seat of most NLEs, and both export to Premiere and Resolve so the handoff is clean either way.
FAQ
Do both work with Claude Code and Cursor?
Yes. invideo lists Claude web and desktop, ChatGPT, Claude Code, Codex, Cursor, VS Code, Gemini CLI, Hermes Agent and OpenClaw. Descript lists Claude desktop, web and Claude Code, ChatGPT, Cursor and any MCP-compatible assistant. Neither needs an API key.
Can I try either for free?
Descript yes - a $0 plan with 60 minutes of media a month, 100 one-time AI credits, 720p watermarked export and no card required. invideo no; its pricing page lists no free plan, so Basic at $9 a seat is the entry point.
Which one can give me a finished MP4?
invideo. Descript’s MCP cannot export a rendered file on request - you publish to a web link and take the signed download URL. Descript’s own documentation flags this as its main gap.
Can the agent spend my credits without limit?
On both, effectively yes, because authentication is your account rather than a scoped key. Descript at least splits the meter into media minutes and AI credits, so an import-heavy run and a processing-heavy run fail differently. Watch unattended loops on either.
Do they replace Premiere or Resolve?
No, and neither claims to. Both are designed to hand off - invideo exports an editable project for Premiere, Final Cut or Resolve, and Descript exports EDL, AAF, SESX, FCPXML, Premiere XML and DaVinci Resolve XML. Treat them as the assembly stage.
What about Runway?
Runway has an official MCP server too, but it is a different job: the agent asks for a generation and gets a render back. It does not hand the agent an editable project you can open and take over, which is the thing invideo and Descript have in common and what makes them comparable here.
Sources
invideo Remote Agents ·
invideo MCP announcement, 30 Sep 2026 ·
invideo pricing, read 28 Oct 2026 ·
Descript MCP ·
Descript MCP documentation ·
Descript pricing, read 28 Oct 2026 ·
descriptinc/descript-mcp