Your video editor, inside your agent
Connect Claude Code, Claude Desktop, Cursor or VS Code to VideoToReels and let it cut, caption, source B-roll and schedule — against your real projects, in your brand style.
claude mcp add --transport http videotoreels https://api.videotoreels.com/mcpThat's the whole setup. Log in through your browser — no API key to paste.
One sentence in, a finished draft out
The agent chains the tools itself — and stops at the approval boundary.
> Take yesterday's podcast recording, pull the strongest 45-second hook,
strip the dead air and the ums, caption it in my brand style,
and put it in the calendar for Thursday morning.
● media.transcript.search found the segment at 12:04–12:51
● edit.cut 3 pieces, snapped to word boundaries
● edit.removeSilences 11 gaps collapsed, 6.2s saved
● edit.removeFillers 9 fillers removed
● captions.generate 142 words, brand style applied
● frame.render 3 frames — checked caption readability
● social.posts.schedule draft saved → preview link (not scheduled yet)Connect your client
The server speaks streamable HTTP at https://api.videotoreels.com/mcp and implements OAuth 2.1 with dynamic client registration and PKCE, so your client registers itself.
claude mcp add --transport http videotoreels https://api.videotoreels.com/mcpThen run /mcp inside Claude Code and pick "Authenticate" — your browser opens, you log in to VideoToReels, and the token is stored for you.
What your agent can reach
67 tools across six groups. Ask any connected client to call capabilities.describe for the live catalog.
Edit the timeline
Add, move, split and delete overlays; re-cut a clip from its source on word boundaries; strip silences and filler words. Every edit is non-destructive — the source media is never rewritten.
- editor.overlay.add
- editor.timeline.applyOps
- edit.cut
- edit.removeSilences
- frame.render
Captions
Generate word-timed captions that read through the cut map, so they stay in sync after a re-cut. Restyle without touching timing, or export SRT/ASS/TXT.
- captions.generate
- captions.updateStyle
- captions.export
B-roll & media
Search licensed stock, insert it with brand treatment applied, search a transcript for the moment a phrase is spoken, and upload local files for posting.
- broll.search
- broll.insert
- media.transcript.search
- media.upload.begin
Publish & schedule
List connected accounts, schedule to them, and publish. Both write paths are preview-then-confirm — the first call returns a draft, never a live post.
- social.posts.schedule
- social.posts.trigger
- schedule.proposeSlots
Growth & strategy
The autonomous side: content streams, strategy horizons, approvals, trend and news signals, workspace memory, and a roster of specialist agents to delegate to.
- growth.draftStrategy
- streams.create
- tasks.assign
- memory.search
- agents.list
Account & workspace
Read the brand contract every edit respects, check the credit balance and monthly AI budget, and list projects.
- brand.get
- credits.get
- budget.status
- capabilities.describe
What it will never do behind your back
Publishing is preview-then-confirm
Scheduling and publishing both refuse to act on the first call. They hand back the resolved caption per platform, the media, the target accounts and a preview link — and only go live when called again with confirm:true. An agent can propose a post; approving it stays yours.
Scoped to you and one workspace
Every tool resolves against your user and your active workspace, the same two-axis scoping the web app enforces. A connection cannot reach another workspace, let alone another account.
Edits are non-destructive
Cuts, silence removal and reorders rewrite a segment map in source time. The media and its word-level transcript are immutable, so every cut is redoable and one transcription stays valid forever.
Revocable and metered
Browser-login tokens are per-client and API keys revoke instantly. AI-heavy tools spend credits, and your agent can read the balance and the monthly cap before it starts anything expensive.
Questions developers ask
- Can an agent post to my social accounts without asking?
- No. Both write paths refuse to act on the first call. social.posts.schedule saves a draft and returns a preview link; social.posts.trigger returns exactly what would go out, to which accounts, with per-platform warnings. Nothing is scheduled or published until the tool is called a second time with confirm:true — which means your agent has to show you the preview and get a yes first.
- Do I need an API key?
- Not for normal use. The server implements OAuth 2.1 with dynamic client registration and PKCE, so your client registers itself and sends you through a browser login. API keys exist for headless environments — CI, shared runners, anywhere a browser cannot open. You can mint and revoke them from Developer → MCP inside the app.
- What can an agent see?
- Only your own data, scoped to your user and your active workspace — the same two-axis scoping the web app enforces. A connection never sees another workspace, and never sees another account.
- Does it cost credits?
- The AI-heavy tools do — transcription, caption generation, B-roll ranking, cover generation. Structural edits, reads, and scheduling are cheap. Your agent can check the balance itself with credits.get and the monthly cap with budget.status before starting anything expensive.
- Are edits destructive?
- No. Cuts, silence removal and reorders rewrite a segment map in source time — the underlying media and its word-level transcript are immutable, so any cut is redoable and one transcription stays valid forever.
- Can I still open the project in the editor afterwards?
- Yes — it is the same project, the same timeline. Agent edits and browser edits go through one shared engine, and writes are guarded by an optimistic version check so the two cannot silently clobber each other.
Point your agent at it
A free account gets you starting credits and the full tool catalog. Connect in one command, revoke whenever you like.