Video editing MCP server for Claude, ChatGPT, and Cursor

Vyra's MCP server connects the AI you already pay for to a full video editor. Claude, ChatGPT, or Cursor gets the actual editing tools, reads your transcripts and scene descriptions, and edits the timeline on Vyra's servers, so no browser tab has to stay open. You describe the video in the chat you already use.

MCP is the Model Context Protocol, an open standard that lets AI assistants call outside tools. We shipped our connector in May 2026 and rebuilt it as a headless server in October 2026. Other editors have added MCP since. The difference is what the agent is allowed to do once it's connected.

What the headless server means

Each agent gets its own editing container on Vyra's servers. Nothing runs in your browser.

• No tab to keep open. The agent creates projects, reads your library, edits, and exports on its own. Your laptop can be closed.
• Edit from your phone. Upload footage straight into the Claude app, ask for the video, and put the phone away while it works.
• Several projects at once. Run one agent across many projects, or several agents in parallel.
• Generated media. The agent can add voiceovers, images, sound effects, and music. Video footage still comes from you or Pexels stock.
• Still a real project. Everything the agent builds opens in the Vyra web and iPhone apps for hand edits.

Which AI clients and models work

Vyra doesn't pick the model. Whatever model your client runs, Claude, GPT, or the one you choose in Cursor, makes the editing decisions and calls Vyra's tools.

How it works

1. Connect your client to https://api.usevyra.com/mcp. It uses Streamable HTTP and OAuth 2.1 with PKCE. A sign-in window opens, you approve, and the client saves the connection. Exact commands are below.
2. Ask the agent to open a project or create one. Nothing has to stay open on your side. The agent edits on Vyra's servers, and the project shows up in the web and iPhone apps whenever you want to look.
3. Ask the agent to look before it cuts. "Summarize the footage and propose a 45-second structure with timestamps."
4. Approve the plan, then go step by step. Rough cut, captions, music, graphics, export.
5. Check with frames. "Capture frames at 0:03, 0:15, and 0:40."

Setup by client

Claude Code. Run these. The second command opens the sign-in window.

claude mcp add --transport http Vyra https://api.usevyra.com/mcp
claude mcp login Vyra

Claude Desktop and claude.ai. Go to Customize > Connectors, click "+ Add", then "Add custom connector". Name it Vyra, enter the URL, and click Continue. Review the detected sign-in, click Continue, choose "Sign in now" under Authentication, and click Add. On Team and Enterprise plans an Owner adds it in Organization settings > Connectors first, then members click Connect. Free plans allow one custom connector.

OpenAI Codex. Run these. The second command opens the sign-in window.

codex mcp add vyra --url https://api.usevyra.com/mcp
codex mcp login vyra

In the ChatGPT desktop app, use Settings > MCP servers > Add server instead. Name it vyra, choose Streamable HTTP, enter the URL, and save. Then click Restart and Authenticate.

ChatGPT on the web. Go to https://chatgpt.com/plugins, click the plus button, then "Add custom MCP server". Name it Vyra, enter the URL under Connection, set authentication to OAuth, accept the warning, and select "Create as a plugin". Install it and sign in.

Cursor. Add this to .cursor/mcp.json (or ~/.cursor/mcp.json for every project), then turn Vyra on in Customize and sign in when asked.

{
  "mcpServers": {
    "Vyra": {
      "url": "https://api.usevyra.com/mcp"
    }
  }
}

Any other client with Streamable HTTP and OAuth 2.1 PKCE works with the same URL. Metadata is at https://api.usevyra.com/.well-known/oauth-authorization-server.

Everything the agent can do

The agent gets 50+ tools. You don't need their names. Describe the result and it picks them.

What you can ask for

Prompts to paste

You are connected to my Vyra project "studio vlog sept". List the assets and summarize the footage. Then propose a 45-second Reel structure with timestamps. Wait for my OK before editing.

Build the rough cut exactly as planned. Remove pauses over half a second and every retake. Keep the last attempt of each line. No captions, no music, straight cuts. Then read the timeline back with start and end times.

For each project in this folder, add 3-5 word phrase captions in the same style, duck the music to 25% under speech, and export a 9:16 version. Tell me when each one is done.

Compared to the usual way

Claude Code and Cursor examples

Developers use Vyra from Claude Code and Cursor to script edits that would take an afternoon by hand.

For every project tagged "client-reels", read the transcript, cut a 30-second vertical version that opens on the strongest line, add 3-5 word captions in the project style, and export 1080x1920. Report the file name and length of each.

Here's my product changelog in CHANGELOG.md. For each of the last 3 releases, find the matching screen recording in my Vyra project, cut a 45-second walkthrough, and put the release title on a title card at the start.

Read the timeline back as a table of clip, start, end, and caption text. Flag any caption longer than 6 words and any clip shorter than half a second.

More in the developer guide and on the Cursor page.

Requirements

• A Vyra account.
• Nothing open on your side. Since October 2026 the server is headless, so the agent works with your laptop closed or from the Claude app on your phone.
• An MCP client from the table above.

Managing sessions

See active MCP sessions and revoke access in Settings > External Agent in the Vyra app. While an agent is connected, an "MCP" tab next to the chat panel shows each tool call as it happens.

Limits

• External models sometimes say a step is done when it isn't. Ask for a timeline read-back.
• Vyra doesn't generate footage. The agent edits what you uploaded.

Troubleshooting

Pricing

3-day free trial, then plans from $24/mo. See /pricing for the full breakdown.

FAQ

Which clients are tested? Claude Desktop and Claude Code the most. claude.ai, ChatGPT, Codex, and Cursor also work.

Which LLMs can I use? Any model your MCP client runs. Vyra provides the editing tools and the footage context, and your model makes the decisions.

Can it edit many videos at once? Yes. Ask for the same treatment across several projects in one conversation, run several agent sessions in parallel, or script it from Claude Code.

Do I need to keep Vyra open in a browser? No. Since October 2026 the server is headless. The agent edits on Vyra's servers, so it keeps going with your laptop closed or your phone off.

Can I use it from my phone? Yes. Add Vyra as a connector in your Claude account and it works in the Claude mobile app. Upload footage in the chat and ask for the video.

How is this different from Descript's or Opus Clip's MCP? Those forward your prompt to the vendor's own assistant. Vyra exposes the editing tools themselves, so your model does the edit and can look at the footage while it works.

Do I need to learn the tool names? No. Describe the result. The full catalog is available through the standard MCP tools/list method, and there's a searchHelpDocs tool for Vyra's help library.

Related

• /answers/video-editors-that-integrate-with-llms
• /claude
• /cursor
• /for/developers
• /docs/mcp
• /best/mcp-servers-for-video-editing