Descript's Model Context Protocol server — drive its transcript-based editor by natural language from Claude, ChatGPT, Codex, and other MCP clients.
Last verified · 2026-11-02 · by Moe Ameen
The Descript MCP is Descript's Model Context Protocol server: a connector that exposes the transcript-based editor so an AI assistant can operate it in plain language. Instead of clicking through the app, you tell an assistant you already use to import a recording, strip the filler, clean the audio, and caption it — and Descript does the work. It's the same editor and the same Underlord AI edits, reached through a chat window rather than the Descript UI. For the scored verdict on the connector itself, see the [Descript MCP review](/reviews/descript-mcp).
Descript demonstrated the workflow across three clients at once. In a free, recorded September 24 public session, three team members each started from an empty project and connected the MCP inside Claude, ChatGPT, and Codex, then tried to complete a real edit within 12 minutes — and the session deliberately walked through where first-time setup breaks. Claude and ChatGPT connect through official connectors in their directories (you sign in with your Descript account, so there's no API token); Codex and any other MCP client connect using Descript's server URL directly.
Over the connection, an assistant can import media, remove filler words and transcript sections, create scenes and highlight reels, apply Studio Sound, add captions, reframe and overlay, translate, read projects back, export transcripts and editing timelines, and publish a composition to a shareable link. Usage draws the AI credits in your existing Descript plan; there's no separate MCP charge.
The boundaries matter for planning. The MCP is scoped to a single Descript drive, it publishes a composition to a link rather than exporting a rendered MP4 on demand, it can't import from a YouTube link, and it edits one project at a time. It makes operating one editor faster — it is not a system that turns a finished cut into a week of platform-native content. Confirm current connectors and limits on descript.com, since they change.
The Descript MCP is the input side of a content operation: an assistant can hand you a clean, captioned master and a highlight reel by chat. What it can't do is generate the formats that aren't just a cut of your recording, or publish anything to a feed. That's the half [Kompozy](/) owns, and the division is sharper than with most tools. Bring a Descript export — the master or a published composition — into Kompozy and it makes the things the MCP has no concept of: [Persona Shorts](/glossary/persona-shorts) and avatar video fronted by your own AI Influencer persona pool, Listicle and Naturalistic videos, face-locked Persona Photos, and brand-exact [Carousel Posts](/glossary/hyperframes) and Quote Graphics. These are net-new generation, not re-cuts of the source.
Then Kompozy covers the part the MCP stops short of: it clips the long recording into vertical [Clipped Shorts](/glossary/clipped-short), reframes each to 9:16, 1:1, and 16:9, spins the transcript into a blog article and an email newsletter, holds every output to one voice with the [Persona Brief](/glossary/persona-brief), and schedules and publishes across the eight social platforms plus blog and email on [Autopilot](/glossary/autopilot) behind a per-post review gate. The clean split: use the MCP to operate Descript by conversation; use Kompozy to generate the formats Descript can't and get all of them onto every feed. See how another editor's MCP fits the same chain on the [DaVinci Resolve 21.1](/ai-tools/davinci-resolve-21-1) page.
It's Descript's Model Context Protocol server — a connector that lets an AI assistant operate Descript's transcript-based editor by natural language. Once connected, you can ask the assistant to import media, remove filler, build highlight reels, apply Studio Sound, add captions, and publish a composition to a shareable link, without opening the app.
Claude and ChatGPT have official Descript connectors in their directories, and you connect by signing in with your Descript account — no API token. Codex and any other MCP client connect using Descript's server URL directly. Behavior in a non-official client depends on how it implements MCP.
Import media, remove filler words or transcript sections, create scenes and highlight reels, apply Studio Sound, add captions, reframe, translate, export transcripts and editing timelines, and publish a composition to a share link. It cannot export a rendered MP4 on demand or import a YouTube link, and it is scoped to one Descript drive at a time.
No separate charge. Connecting an assistant is free, and the edits you trigger draw the AI credits in your existing Descript plan. Automated editing can use credits faster than clicking through the app, and running out of media minutes or AI credits returns a 402 error on the request.
Use the MCP to operate Descript by chat and produce a clean master. Then bring that export into Kompozy to generate the formats Descript cannot — persona and avatar video, carousels, quote graphics, a blog, and a newsletter — clip it into per-platform shorts, and publish everything across the eight social platforms plus blog and email. The MCP edits one project; Kompozy generates and distributes the content around it.