Make Documentaries
With Your AI Agent
Claude, Claude Code, Codex, OpenClaw, Hermes, Grok and ChatGPT can write a great script. On their own they make, at best, a 10-second clip. Connect Cliptude and the same agent directs a finished YouTube documentary up to an hour long.
Your agent picks the footage under every line, the maps, charts and titles, the cuts and the sound. Cliptude narrates, sources real footage, builds the graphics and renders the video.
In Claude Code
claude mcp add --transport http cliptude https://mcp.cliptude.com/mcp
Then ask
“Make a 10-minute history documentary about how TSMC became the world's chip foundry. Direct every shot with real footage, put the key numbers on screen as charts, and show me the plan before you render.”
You get
A rendered MP4 with narration, captions, footage from stock libraries and public archives, maps, data charts, animated title cards and transitions, plus a shot-by-shot review you can change in the editor.
Agents Make Clips. Cliptude Makes the Film.
Every agent can now generate a few seconds of video. A YouTube documentary needs ten minutes to an hour of it, cut to the narration.
Generated clips
Grok Imagine, OpenClaw's video_generate tool and the text-to-video models behind most MCP servers make one short shot per prompt. Stitching 120 of them into a story, with narration and the right picture on every line, is still your job.
Code from scratch
Claude Code and Codex can write a Remotion project. That works for a product promo. For a documentary they also need a voice, a footage search, licensing, maps, charts and a renderer, and they rebuild all of it for every video.
A production pipeline
Cliptude gives the agent a studio it can direct: narration with word timings, real footage and archive search, nearly 200 animated boards, maps and charts, transitions, sound and rendering. The agent makes the creative calls; Cliptude makes the film.
Three Ways to Connect an Agent
Same tools, same plan format, same results. Pick whatever your agent speaks.
For Claude, ChatGPT, Grok, Codex, Cursor, Hermes, OpenClaw and any host that supports remote MCP servers. Sign in with OAuth; no keys to paste.
https://mcp.cliptude.com/mcp
MCP setup →
For coding agents and terminals without MCP. The agent runs the same steps as shell commands and reads JSON back.
npm i -g @cliptude/cli
cliptude auth login
CLI reference →
For your own agents, workflows and automations. Create a job from a prompt or script, poll it, fetch the video.
curl -X POST https://cliptude.com/api/v1/jobs \
-H "Authorization: Bearer clp_…" \
-H "Content-Type: application/json" \
-d '{"template_id": "uuid",
"title": "The rise of TSMC",
"prompt": "how TSMC came to make most advanced chips"}'
API docs →
Add the Cliptude skill so your agent knows the plan format, the graphics catalog and what makes a good documentary:
npx skills add cliptude/cliptude
Make Videos With Your Agent
Setup takes a minute in each of them. The model you pick changes the taste and cost of the plan, not the workflow.
Make videos with Claude and Claude Code
In claude.ai or Claude Desktop: Customize → Connectors → Add custom connector, paste the URL and sign in. In Claude Code, one command:
claude mcp add --transport http cliptude https://mcp.cliptude.com/mcp
How to make a documentary with Claude →
Make AI avatar videos with Claude →
Make videos with Codex
Works in the Codex CLI, the IDE extension and the ChatGPT desktop app. Codex can also drive the CLI directly.
codex mcp add cliptude --url https://mcp.cliptude.com/mcp
codex mcp login cliptude
Codex setup →
OpenClaw video generation, full length
OpenClaw's video_generate makes single clips. Add Cliptude as an MCP server and the skill above, and your OpenClaw agent can turn a topic into a finished YouTube documentary.
OpenClaw setup →Hermes Agent video skill
Hermes reads the same SKILL.md format, so the Cliptude skill installs as is. Connect the MCP server and Hermes can plan, build and render videos on a schedule.
Hermes setup →Make long videos with Grok
Grok Imagine makes clips of a few seconds. Add Cliptude at grok.com/connectors → New Connector → Custom, paste the URL and sign in. Grok then directs full-length videos.
Grok setup →ChatGPT, Cursor and other MCP hosts
ChatGPT custom MCP apps, Cursor, VS Code, Windsurf and any client that speaks Streamable HTTP connect with the same URL and OAuth sign-in.
Other hosts →How an Agent Directs a Documentary
The agent makes the editorial calls. Cliptude checks them and does the production.
Script and narration
The agent writes the script, or you bring your own or upload a voiceover. Cliptude records the narration and returns every word with its timing, so the agent can cut to the exact syllable.
A shot-by-shot plan
The agent plans the picture under each sentence, with search terms, the era and what must be visible, plus every name, number, map, chart and title card, the transitions and the sound hits.
Checked before it is built
Cliptude checks the plan the way an editor would: empty boards, slow openings, long stretches with nothing on screen, pictures held too long, missing search backups. The agent fixes them before a credit is spent on the build.
Built, reviewed, rendered
Cliptude sources the footage, builds the graphics and assembles the timeline. The agent reads a review of every shot, swaps what's wrong, and renders. You can open the project in the browser editor at any point.
A Brief to Paste Into Any Agent
Change the topic and length (up to an hour). It works the same in Claude, Codex, Grok, Hermes or OpenClaw.
Connect the Cliptude MCP server at https://mcp.cliptude.com/mcp and sign in. If you can't use MCP, install the CLI with `npm i -g @cliptude/cli` and run `cliptude auth login`.
Make a [length, up to 60 minutes] YouTube documentary about [your topic], in the History look.
0. Set up a template with me using Cliptude's pickers, one at a time: voice, music, background and style.
1. Write a script with a strong cold open, then record the narration.
2. Direct the whole film: real footage under every sentence (with backup search terms and the era), names, dates and numbers on screen, maps for places, charts for figures, and a few transitions at the chapter turns.
3. Fix everything check_plan reports, then show me the plan and the review card with its cost before you build.
4. After the build, review every shot, replace anything off-topic, render, and show me the publish kit.
The full walkthrough, with what to check at each step: How to make documentaries on Claude.
What Your Agent Can Make
The same agent workflow covers every format the plan supports.
History documentaries
Archive footage, period photos, maps of empires and wars, dated title cards.
True crime and business stories
Timelines, quote cards, company logos, revenue charts and evidence boards.
Top 10 countdowns
Ranked entries with their own titles, footage and stats.
Avatar explainers
An AI presenter on screen where the agent places them, with B-roll in between.
Frequently Asked Questions
Making videos and documentaries with AI agents.
Can Claude make a video?
Can Codex make videos?
codex mcp add cliptude --url https://mcp.cliptude.com/mcp, sign in with codex mcp login cliptude, and ask Codex to make the video. Codex can also use the Cliptude CLI or REST API without MCP.
Can Grok make long videos?
How is this different from OpenClaw's video_generate tool?
How long can an agent-made video be?
Do I need a template first?
Do I need to write code?
Can I review the agent's plan before it renders?
Is an agent-made documentary allowed on YouTube?
Give your agent
a film studio
Connect Cliptude to Claude, Codex, OpenClaw, Hermes, Grok or ChatGPT and make your first agent-directed documentary today.