MCP server
Make films from the AI chat you already use
Connect VidSwarm to Claude, ChatGPT, Cursor or Codex once. Ask for a film, follow it, review notes and render, without leaving the chat. Anything that spends credits waits for your yes.
Server address
https://vidswarm.app/api/mcp
An illustration with VidSwarm’s real tool names; the channel and film are examples. The windows are simplified sketches, not the apps themselves.
One server, any assistant
Every assistant talks to the same VidSwarm MCP server, and the films land in the same dashboard.
- Claude
- Cursor
- Codex
- ChatGPT
Recent films
- The Silk RoadIn review
- Why Rome Built Roads FirstRendering
- The Lost Fleet of Kublai KhanPublished
Example titles
How it works
What happens between your message and a finished film, step by step.
- Step 01
Ask in your assistant
Say what the film is about and for which channel, in plain words, in the chat you already use.
You, in your assistant
“Make a 10-minute video on the history of the Silk Road for my history channel.”
- Step 02
Your channel is picked
The assistant lists your channels and takes the one you named. The film follows that channel’s look, voice, music and language.
list_channels✓Channel
Ancient Routes
History · English · its look, voice and music
- Step 03
The film is created for free
create_film starts the film and plans it. Planning is free; nothing is charged at this point.
create_film✓Planning is free✓Nothing charged✓ - Step 04
You see the price first
get_film returns the plan with its price in credits. Credits are reserved only when the storyboard is approved, and the assistant may approve it only after your explicit yes.
get_film · plan and price✓Price of this film700 creditsReservedonly after your yes - Step 05
VidSwarm makes the film
Research, script, voice, footage, graphics and the cut run in VidSwarm. render_status shows the progress; a render reserves its price when it starts.
Storyboard approved by you✓Research, voice, footage, cut✓Rendering42% - Step 06
Review it in the editor
The film waits in your dashboard. Open it in the editor to review, change scenes, render the changes and publish.
Ready to reviewin review16:9The Silk Road · 10 minvidswarm.app/dashboard/…Open in editor →
You, in your assistant
“Make a 10-minute video on the history of the Silk Road for my history channel.”
Your channel’s style comes along
The assistant doesn’t need to describe your look. Every film follows the channel it is made for: its colours and fonts, voice, music, language and pacing, set once in the channel’s settings.
Set up a channel in the studio, or let the assistant create one with save_channel. Channel look
- Look
- Fonts
- Display serif · Sans
- Voice
- Calm narrator, slightly slow
- Music
- Low strings, quiet
- Language
- English, dubbed into Spanish
- Pacing
- Slow build, question hook
- Strategy
- Three reference videos its films and thumbnails follow
14 tools, nothing more
Everything the server lets an assistant do. Only two of them can reserve credits, and both say so first.
list_channelsYour channels with niche, language and their YouTube connection.0 creditssave_channelCreates or updates a channel (connecting YouTube itself happens in the studio).0 creditscreate_filmStarts a film from an idea. Planning is free; nothing is charged.0 creditsget_filmA film’s brief, plan with its price, approvals, script and what it waits for.0 creditsget_taskThe next step waiting for your own agent: plan, script, upload texts, Shorts, an approval.0 creditssubmit_resultHands back your agent’s answer; VidSwarm checks it like its own model’s.Approving the storyboard reserves the film’s credits, only after your yeslist_filmsYour films with open review notes, the latest version and whether one is rendering.0 creditsfilm_notesThe review notes of a film: time, scene, the region drawn on the frame, status.0 creditsadd_noteAdds a review note at a time or range of the film.0 creditsnote_updateMarks a note fixed, still there, resolved or open, or changes its text.0 creditsrenderRenders the changed scenes, the whole film, or a free 720p preview.Reserves the render’s price (the preview is free)render_statusVersions and previews: status, progress, time left, price.0 creditsthumbnail_briefA Thumbnail Studio sketch as a written brief, with your marks.0 creditsbalanceCredits left in the workspace.0 credits
Connect VidSwarm to your assistant
One server address for every assistant: sign in with VidSwarm in Claude, ChatGPT and Claude Code, or use an API key in Cursor and Codex. Ask it to list your channels to check the connection.
Server address (Streamable HTTP)
https://vidswarm.app/api/mcp
- 1In Claude (claude.ai or the Claude app) open Settings › Connectors › Add custom connector.
- 2Paste the server address below and click Connect.
- 3Sign in with VidSwarm, pick the workspace and approve. You can untick “Write films” and “Spend minutes”; reading is always on.
https://vidswarm.app/api/mcp
Or let your agent set it up
Paste this into Claude Code, Codex or Cursor. It asks you for the key and never wants it pasted into the chat.
Set up VidSwarm (https://vidswarm.app) for me in this project.
1. Ask me to create an API key at https://vidswarm.app/dashboard/developers/api-keys ("Create key") and to put it into the environment variable VIDSWARM_API_KEY myself. Never ask me to paste the key into this chat, and never write it into a file in the repository.
2. Connect VidSwarm's MCP server (Streamable HTTP) at https://vidswarm.app/api/mcp with the header "Authorization: Bearer $VIDSWARM_API_KEY".
- Claude Code: claude mcp add --transport http vidswarm https://vidswarm.app/api/mcp --header "Authorization: Bearer $VIDSWARM_API_KEY"
- Codex: in ~/.codex/config.toml add [mcp_servers.vidswarm] with url = "https://vidswarm.app/api/mcp" and bearer_token_env_var = "VIDSWARM_API_KEY"
3. Check the connection by calling the MCP tool "balance" and tell me the credits left.
4. The REST API is at https://vidswarm.app/api/v1 with the same header; the endpoints are documented at https://vidswarm.app/docs/api.Questions about MCP
The Model Context Protocol is an open standard that lets AI assistants call tools in other apps. VidSwarm runs an MCP server, so your assistant can make, review and render films in your VidSwarm account.
Claude (claude.ai and the Claude app), ChatGPT in developer mode and Claude Code connect with “Sign in with VidSwarm”. Cursor and Codex connect with an API key.
No. In Claude and ChatGPT you paste the server address and sign in; in Cursor, Codex and Claude Code it is one command or a few lines of settings, shown above.
Anyone with a VidSwarm account. Connecting an app and creating an API key are for workspace owners; both act in the workspace they were made for.
List and save channels, create films, read where a film stands, handle review notes, render, and read your credit balance. When you approve an app you can untick “Write films” and “Spend minutes”; reading is always on.
Creating and planning a film is free. Credits are reserved when the storyboard is approved, which the assistant may only do after your explicit yes, and when a render starts. Untick “Spend minutes” when you approve an app and it cannot spend at all.
You can leave and update review notes, render changed scenes, the whole film or a free preview, and follow the render. Changing scenes, the final review and publishing happen in the editor, where the film waits for you.
Signing in never shares your password with the assistant, and an API key is never needed in the chat or in a project file. A connected app has the same tools, permissions and rate limit as an API key and stays signed in until you disconnect it under Developers › Connected apps; a key can be disabled or deleted at any time.