What is a video editing MCP server?

It is a service that exposes video editing operations as tools an AI assistant can call through the Model Context Protocol. The assistant decides what to do; the server holds the footage, applies the edits and renders the video. Valmera's server edits footage you upload: it reads the word-level transcript and frames, cuts, captions, reframes, adds music and b-roll, renders previews and exports an MP4.

Which AI assistants can use Valmera?

Any MCP client that supports remote Streamable HTTP servers: Claude (claude.ai, Desktop and Claude Code), ChatGPT, Cursor, VS Code with GitHub Copilot, Codex, Windsurf, Zed, Cline, Continue, Goose, LibreChat, Gemini CLI and the MCP Inspector. Clients that only run local stdio servers can connect through the mcp-remote bridge.

Can the assistant export the final MP4 over MCP?

Yes. When you approve a preview, the assistant calls export_final with the project ID and the reviewed edit version, follows the job with wait_for_job, then gets the file with download_url(kind="final"). The export is a full-resolution H.264 MP4 rendered from your original upload and ends with a brief Valmera end card. You can also export from Studio.

Do I need a subscription?

Creating an account and uploading are free. Editing requires a subscription, and MCP access is included with eligible Valmera subscriptions; see the plans at valmera.io/subscribe. Your AI assistant is billed separately by its own provider.

Does the assistant actually see the video?

Yes. Tools return transcript text, analysis and sampled frames from the source or from the rendered edit, and watch_video hands over the MP4 itself for models that accept video. That is how the assistant checks its own work before it reports back.

How does my footage get to Valmera?

Upload it in Studio, or let a client with a shell (Claude Code, Cursor, Codex) upload a local file: upload_start returns a presigned URL, the client PUTs the bytes, and upload_finish starts the analysis. Files can be MP4, MOV, M4V, MKV or WebM, up to 100 GB and 3 hours. The assistant can also import clips, music and images from a link with fetch_url.

Is my original file changed?

No. Every edit writes a new version of a non-destructive edit decision list. The upload is never modified, any cut can be restored, and the final export renders from the untouched original.

The MCP server is the API. It speaks JSON-RPC 2.0 over Streamable HTTP, so you can call it from your own code with a bearer token or OAuth, with or without an LLM. The video editing API page has curl, TypeScript and Python examples.

Video editing MCP server

Valmera is a hosted MCP server that gives Claude, ChatGPT, Cursor, VS Code, Codex and any other MCP client a real video editor. Your assistant reads the word-level transcript and the frames of footage you uploaded, cuts silences and filler words, adds word-timed captions, reframes to 9:16, adds b-roll, music and sound effects, renders a preview it can watch, and exports a full-quality MP4 from your original file.

It edits footage you already have. It is not a text-to-video generator.

ENDPOINT

https://valmera.io/mcp/server

Transport: Streamable HTTP (JSON-RPC 2.0)
Auth: OAuth 2.1 with dynamic client registration and PKCE (S256), scope valmera.edit, or a bearer token
Tools: 143 tools: 129 editing tools in 9 groups plus 14 session tools (live server card, October 2026)
Registry: io.valmera/video-editor in the official MCP Registry

Connect your assistant → · Live server card · Docs on GitHub

Connect in one minute

Most clients need only the URL: they read Valmera's sign-in details, register themselves and open a Valmera login. There is no client ID, secret or API key to create.

Claude (claude.ai and Claude Desktop)

Customize → Connectors → + Add → Add custom connector. Name it Valmera, paste the URL and continue. If Claude asks how it should identify itself, choose Register automatically. Sign in to Valmera and allow access.

SERVER URL

https://valmera.io/mcp/server

Claude guide →

Claude Code

Run the command, type /mcp inside Claude Code and choose Authenticate.

TERMINAL

claude mcp add --transport http valmera https://valmera.io/mcp/server
# then, inside Claude Code:
/mcp

Claude Code guide →

Cursor

Paste the entry, or use the one-click install. Cursor opens the Valmera sign-in itself.

~/.cursor/mcp.json

{
  "mcpServers": {
    "valmera": { "url": "https://valmera.io/mcp/server" }
  }
}

Add to Cursor (one click) · Cursor guide →

VS Code (GitHub Copilot agent mode)

The top-level key is servers, not mcpServers. VS Code opens a browser to sign in on first use.

.vscode/mcp.json

{
  "servers": {
    "valmera": { "type": "http", "url": "https://valmera.io/mcp/server" }
  }
}

VS Code notes →

ChatGPT

Open chatgpt.com/plugins → + → Create custom MCP server. Name it Valmera, paste the URL under Connection, keep OAuth, accept the notice and create it. In a new chat, type @ and choose Valmera. If the create option is missing, turn on Developer mode in Settings → Security and login.

SERVER URL

https://valmera.io/mcp/server

ChatGPT guide →

Codex CLI

The login command opens the Valmera sign-in in your browser.

TERMINAL

codex mcp add valmera --url https://valmera.io/mcp/server
codex mcp login valmera

Codex notes →

Windsurf

Cascade uses at most 100 tools at a time across all servers, and Valmera exposes 143. List the tools you don't need in the server's disabledTools array.

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "valmera": { "serverUrl": "https://valmera.io/mcp/server" }
  }
}

Windsurf notes →

Any other client

Add https://valmera.io/mcp/server as a remote Streamable HTTP server. A client without OAuth can send Authorization: Bearer vlm_mcp_… with a private token you create on the connect page. A client that only runs local stdio servers can use the mcp-remote bridge:

STDIO BRIDGE

{
  "mcpServers": {
    "valmera": { "command": "npx", "args": ["-y", "mcp-remote", "https://valmera.io/mcp/server"] }
  }
}

Zed, Cline, Continue, Goose, LibreChat, Gemini CLI and the MCP Inspector each have their exact config on the setup page. Once connected, send “List my Valmera projects.” An empty list still means the connection works.

What your assistant can do

You describe the outcome; the assistant picks the tools. These are the ones it reaches for most. The tool reference covers all 143.

You ask for	Tools the assistant uses
“Remove the dead air and the ums”	find_silences, cut_silences, remove_filler_words, get_kept_transcript
“Cut the part where I talk about pricing”	search_transcript, get_words, cut_range, restore_range
“Add bold word-by-word captions”	add_captions, set_caption_style, set_caption_fixes, audit_captions
“Make it vertical and keep my face centered”	auto_reframe, set_frame, add_aspect_shift
“Turn this podcast into five shorts”	make_shorts, shorts_status, open_short, apply_short_edit_batches
“Add b-roll where I mention the product”	research_broll, search_stock, add_stock_media, insert_media
“Punch in on the important lines”	suggest_emphasis, punch_in_on_emphasis, add_zoom, add_zoom_path
“Put this song under my voice and duck it”	find_song, fetch_url, add_music, set_music_fit, set_master_loudness
“Cut the montage to the beat”	get_audio_analysis, beat_align_cuts
“Add a whoosh on each transition”	search_sfx, audition_sfx_candidates, add_sfx
“Put a title card at the start”	add_title_card, add_text, add_kinetic_text, add_text_behind
“Record a demo of my website”	record_website, record_website_demo, enhance_cursor, showcase_demo
“Blur the license plate” / “remove the burned-in subtitles”	blur_region, find_burned_text, erase_burned_text, erase_region
“Grade it warmer”	set_color_grade, apply_look, set_grade_custom
“Show me what you changed”	render_preview, wait_for_job, watch_video, look_at
“Export the final video”	export_final, wait_for_job, download_url

How an edit works

Upload. Upload in Studio, or let a client with a shell upload a local file with upload_start and upload_finish. MP4, MOV, M4V, MKV or WebM, up to 100 GB and 3 hours. fetch_url imports clips, music and images from a link.
Index. Valmera builds a word-level transcript with speaker labels and detects shots, silences and on-screen text. index_status reports progress.
Edit. Every tool call writes a new version of a non-destructive edit decision list. The original is never modified, and restore_range brings any cut back.
Review. render_preview renders the edit, and watch_video and look_at give the assistant the frames it produced so it can fix problems before it replies.
Export. export_final renders a full-resolution H.264 MP4 from the original file for the version you approved.

Slow work (renders, pixel repair, website recording) returns a job ID instead of a fake completion. The assistant calls wait_for_job until the job finishes. Final delivery looks like this:

EXPORT FLOW

export_final(project_id=4127, edl_version=7)
  → export of EDL v7 is job 9120
wait_for_job(job_id=9120)        # call again while it is still running
  → done
download_url(project_id=4127, kind="final", edl_version=7)
  → temporary link to the MP4

The exported MP4 ends with a brief Valmera end card. Studio's Export button does the same job if you would rather start it yourself.

Example prompts

Say who the video is for, what must stay and how you will judge it. Paste one of these to start:

Open my latest project. Remove dead air and filler words but keep natural pauses, add clean white captions with the active word highlighted, reframe to 9:16, render a preview and tell me what you changed.

This is a 70-minute two-person podcast. Find the 5 strongest self-contained stories, make each one a vertical short with captions and a hook title, and keep whoever is speaking in frame.

Record a 30-second demo of https://example.com: scroll the hero, click Pricing and zoom on the main button. Then add captions and put the music track I uploaded underneath.

Restore the pause before the final answer; it felt too abrupt. Keep everything else and render a new preview.

Export the version I just approved as a full-quality MP4 and give me the download link.

More in the prompt library and the 45 MCP prompts on GitHub.

How Valmera compares to other video MCP servers

Video MCP servers do different jobs. Pick by the output you need:

Kind of server	Examples	Best for
Hosted editor for real footage	Valmera	Editing recordings end to end from a chat: transcript- and frame-aware cuts, captions, reframing, b-roll, music, previews and MP4 export
Editing inside another editor	Descript, Rendley	Teams already working in that product's projects and editor
Clipping and publishing	Reap, OpenClip, Riverside	Clips from long videos and scheduling posts to connected social accounts
Desktop editor bridges	DaVinci Resolve MCP	Driving a local Resolve project that you finish by hand
FFmpeg wrappers	video-audio-mcp and similar	Local, deterministic conversions when you already know the timestamps
Generation	Runway	Generating new shots from prompts and reference images

The full comparison covers each server's connection, documented operations and delivery, with dated sources.

Use it from your own code

The MCP server is Valmera's API. It speaks JSON-RPC 2.0 over Streamable HTTP, so a script, a pipeline or your own agent can call the same tools with a bearer token, no LLM required. The video editing API guide has curl, TypeScript and Python examples, the job model and the upload and export flow.

Limits

It edits footage you provide. It does not generate a whole video from a prompt.
Captions are burned in. There is no SRT or VTT import or export.
No direct publishing to YouTube or TikTok: download the MP4 and post it.
No team seats or collaboration yet.
No custom font uploads, true crossfade or dissolve transitions, or motion-tracked stickers.
No audio denoise ("studio sound"), AI music generation or mood-based music library: music comes from your own file or a song you name.
Over MCP, your assistant edits with the tools directly. It can't hand the job to Valmera's in-house agent.
One editor per project at a time: editing the same project from Studio and over MCP at once is refused in both directions.
English is the best-tested transcription language.

Every MCP guide

Connect your assistant — Pick your app, connect, and see whether Valmera has received its first request.
Setup for every MCP client — Config file, exact JSON and auth path for each client, plus OAuth, token and troubleshooting details.
Tool reference — Every tool grouped by job, with what each one does.
Claude — Custom connector setup for claude.ai and Claude Desktop.
Claude Code — One command, scopes, a team .mcp.json and a session from local file to MP4.
ChatGPT — Add Valmera as a custom MCP server and edit from a chat.
Cursor — mcp.json, one-click install and screen-recording edits from the editor.
What is MCP video editing? — How an assistant, an MCP server and a video engine split the work.
Video editing MCP servers compared — Valmera, Descript, Reap, OpenClip, Resolve, Runway, Riverside and more.
Video editing API for developers — Call the same server from code: curl, TypeScript and Python.