60+ tools · Claude Code · Codex · Cursor · any MCP client

Your coding agent, now a creative studio.

Cleus gives Claude Code, Codex and Cursor video, images, voices, dubbing, music, sound effects, 3D, roleplay characters, long-form stories and uncensored chat — plus 1,400+ more models — as tools they call on their own while they build. One URL, one API key.

https://api.cleus.ai/mcp
tools
60+tools
video models
8video models
image models
8image models
more models on demand
1,400+more models on demand
>

Scripted demo of a typical session: plan → generate → poll → download → assemble.

Capabilities

Everything Cleus can make, as tools

Pick a category to see the models, what they’re best at, and prompts to try.

Shots, edits, swaps and dubs — with sound

8 video models behind one tool, from quick 480p drafts to 30-second 1080p shots with spoken dialogue. Animate any image between a start and end frame, restyle footage you already have, put anyone into a clip's exact moves — then dub it, reframe it and add sound.

  • Text-to-video or image-to-video: start + end frames, 1–30 seconds, 480p to 1080p, landscape, vertical, square or cinematic 21:9.
  • Native sound and dialogue on Quality+, Kling 4.0, Seedance 2.5, Veo 3, Sora 2 and Grok Imagine — write the lines in quotes.
  • Kling 4.0 edit: “turn @Video1 into claymation”, swap the setting, outfits or lighting while keeping the motion. Reference mode follows a clip’s camera and your reference images.
  • Swap into video: the person or mascot in a photo performs the exact moves of any clip, original audio kept — preview the first 5 s cheaply, then render in pro.
  • Extend Veo shots ~7 s at a time; background music, seeds and negative prompts on Uncensored+.
  • Talking videos: any portrait (or your Avatar) speaks your script in one of 21 presenter voices, or your own audio, lip-synced for up to 60 s.
  • Dub any video into 30+ languages in the speakers’ own voices, or lip sync a clip to new audio.
  • Reframe 16:9 ↔ 9:16 ↔ 1:1 by generating the missing edges instead of cropping, and add synced sound effects to silent clips — made from the picture.
  • Upscale to 4k, cut out the background, or make it slow motion.
  • Faceless videos: one idea in, narrated and captioned shorts out — 8 formats, 18 looks, up to 4 parts that continue each other.
  • Remix a viral video: recreate its format with your topic, swap yourself in, or restyle it. Video analysis breaks any clip into shots, cuts and a timed transcript first.
  • The Imagine Agent directs whole productions on its own: script, shots, voice, music, review and the final cut — capped by your budget.
Quality+

Best all-rounder: 2–30 s, up to 1080p, native sound and dialogue, start + end frames

2–30 s · 480p / 720p / 1080p · model "quality"

Uncensored+

No filter, allows explicit (adults only), 5 or 10 s, optional background music

5 or 10 s · 480p / 720p / 1080p · model "uncensored"

Kling 4.0

Cinematic motion with sound, 3–30 s, 720p or 1080p, start + end frames

3–30 s · 720p / 1080p · model "kling"

Seedance 2.5

Top detail, multi-shot scenes and dialogue, 4–15 s

4–15 s · 480p / 720p · model "seedance"

Seedance 2.0

Reliable motion with native audio, 3–12 s, 720p, start + end frames

3–12 s · 720p · model "seedance-2"

Veo 3

Realistic footage with sound, 4 or 8 s; can be extended with cleus_extend_video

4 or 8 s · 720p · model "veo"

Sora 2

Storytelling shots with sound, 4 or 8 s

4 or 8 s · 720p · model "sora"

Grok Imagine Video

Real public figures, 1–15 s, up to 1080p, with sound

1–15 s · 480p / 720p / 1080p · model "grok-imagine"

Talking video

A portrait speaks a script or your audio — Standard, Pro or Ultra

up to 60 s · 21 voices

Faceless video

Script, visuals, narration, music and captions from one idea

8 formats · 18 looks · 9:16 / 16:9 / 1:1

Dubbing

Translated speech in the original speakers’ voices

30+ languages

Reframe

Any aspect ratio by generating the edges, no cropping

9:16 / 16:9 / 1:1 / 4:3 …

Remix

Clone a viral video’s format, swap in, or restyle

recreate · swap · restyle

Imagine Agent

Plans, generates, reviews and edits multi-shot productions

ads · music videos · shorts · remakes

Ask your agent

“Animate public/hero.png into a slow 8-second dolly-in with Cleus Kling at 1080p and save it to public/hero.mp4.”
“Make a 20-second vertical Cleus video of a street musician singing “we were young once”, with sound.”
“Turn demo/screen-recording.mp4 into a hand-drawn anime version with Cleus, same camera move.”
“Dub demo/product.mp4 into Spanish and German with Cleus and reframe both to 9:16.”
“Make a 60-second faceless “what if” video about the moon disappearing with Cleus, toy diorama look.”
“Have the Cleus Imagine Agent make a 30-second ad for my coffee brand from these product photos.”

How it works

Agents chain the tools for you

  1. 1

    You ask

    Mention Cleus in a normal prompt — or run a workflow like /mcp__cleus__trailer in Claude Code.

  2. 2

    Your agent plans

    It reads the 60+ tool descriptions, picks models, uploads local files and chains calls together.

  3. 3

    Cleus generates

    Jobs run on the same backend as cleus.ai. Slow ones (videos, songs) return an id your agent polls with cleus_get_result.

  4. 4

    Files land in your repo

    Results are hosted URLs; your agent downloads them, cuts videos with ffmpeg, wires assets into code.

Long jobs never time out

Every call waits ~45 s, then returns status: processing and an id. cleus_get_result picks it up — even in a later session. Clients with long timeouts can pass wait_seconds up to 240.

Local files in, files out

cleus_upload_file takes base64 or hands your agent a curl upload URL for big videos. Outputs are hosted URLs your agent downloads into the repo.

Same account as the app

Generations, stories and character chats appear in your Cleus history, Story Studio and DMs — start on the CLI, finish on your phone.

Live previews in the chat

In clients that support MCP Apps, results appear as an inline player or gallery that fills in on its own, with download and one-click follow-up buttons. cleus_upload_widget opens an upload box for your photos and videos.

Batches and chains

cleus_generate_batch starts up to 12 jobs at once, cleus_get_result waits on all of them and cleus_show_results puts them in one gallery. Any file input takes a job id (<id>:2 for the second result) or an @handle, so results flow into the next step.

Check before you run

Add get_cost: true to any paid tool and it tells your agent the exact cost and whether your balance covers it — nothing runs. cleus_transactions shows what you spent per day.

Workflows

Whole productions in one prompt

Each workflow is also an MCP prompt: in Claude Code type /mcp__cleus__trailer, /mcp__cleus__audiobook and so on. Other agents load the same plans with cleus_workflows. Click a step to see the exact tool call.

5 tool calls~8 min

>Make a 15-second launch trailer for my neon samurai roguelike with music, and save it to marketing/trailer.mp4.

  1. {
      "prompt": "Neon samurai on a rain-soaked rooftop, cinematic key art",
      "model": "pro",
      "n": 3,
      "aspect_ratio": "16:9"
    }

Reference

All 60+ tools

Generated from the live server — exactly the names and parameters your agent sees.

Setup

Connected in 2 minutes

  1. 1Get an API key

    Open cleus.ai/developers, sign in and click Create key. Copy the clsk_live_… key right away — it’s only shown once.

    MCP uses your Cleus account, so you need an active subscription.

  2. 2Connect your agent

    Replace clsk_live_YOUR_KEY with your key:

    Run this once in any terminal. --scope user makes Cleus available in every project.

    Terminal
    claude mcp add --transport http --scope user cleus https://api.cleus.ai/mcp \
      --header "Authorization: Bearer clsk_live_YOUR_KEY"

    Start claude and type /mcp. You should see cleus · connected with 60+ tools. Videos can take a few minutes; give tools more room with MCP_TOOL_TIMEOUT=300000 claude.

    To share it with a team, commit a .mcp.json in the repo root and have everyone set CLEUS_API_KEY in their shell:

    .mcp.json
    {
      "mcpServers": {
        "cleus": {
          "type": "http",
          "url": "https://api.cleus.ai/mcp",
          "headers": { "Authorization": "Bearer ${CLEUS_API_KEY}" }
        }
      }
    }
  3. 3Ask for something

    Mention Cleus and your agent picks the tools. Start small:

    “Use Cleus to make a 16:9 hero image for this landing page and save it to public/hero.png.”
    “List the Cleus video models and tell me which one fits a 20-second vertical clip with dialogue.”
    “Run /mcp__cleus__game-assets for 6 fantasy potion icons.”

FAQ

Questions & troubleshooting

›What do I need?

A Cleus subscription and an API key from cleus.ai/developers. Add the key to your client — setup takes about two minutes.

›Which clients work?

Claude Code, Codex, Cursor, Claude Desktop, VS Code (Copilot agent mode), Windsurf, Zed, Cline, Goose, Gemini CLI, OpenCode and anything else that speaks MCP over Streamable HTTP — or stdio through the mcp-remote bridge.

›My video came back as “processing”.

That’s normal: videos take 1–5 minutes and songs 1–3. Each call waits about 45 s, then hands back a job id; your agent calls cleus_get_result until it’s done. Never re-run a processing job — that would start it twice. If your client allows long tool calls, pass wait_seconds (up to 240).

›How do I use my own files?

Any public URL works as an input. For local files your agent calls cleus_upload_file: small files go up as base64, big ones (videos) get a one-hour upload URL and a ready curl command — Claude Code and Codex run it for you. In chat apps that support MCP Apps, cleus_upload_widget shows an upload box instead.

›Do results show up inside the chat?

Yes, in clients that support MCP Apps: images, videos, audio, 3D models and SVGs appear as live previews that update while a job runs, with download buttons and one-click follow-ups. Other clients get the same results as text and links.

›Can I use a model Cleus doesn’t list?

Probably. cleus_find_models searches 1,400+ more image, video, music, speech, audio and 3D models and shows each one’s inputs; cleus_run_model runs it, with the cost known before it starts. Ask for a model by name or describe the job.

›Where do results go?

Cleus returns hosted URLs. Ask your agent to save them in your project (“put it in assets/”). Download right away: links from some models expire within hours. Everything also shows up in your Cleus history (cleus_list_generations).

›Is it uncensored?

Chat, Story mode, characters, Uncensored+ images and Uncensored+ video don’t filter adult content for adults. Some partner models (Kling, Veo, Sora, Grok, Pro+ and the 1,400+ catalog models) keep their own provider filters. Content involving minors, real private people without consent, and illegal material is never allowed.

›Can I call the same tools from my own code?

Yes. Every MCP tool is also a REST endpoint at api.cleus.ai/v1 — /videos/generations, /videos/dub, /audio/sound-effects, /images/3d, /models/run, /characters/{id}/messages, /stories and more — with the same key, plus an OpenAPI spec at api.cleus.ai/v1/openapi.json. See cleus.ai/developers.

›Do characters and stories sync with the app?

Yes. A character chat over MCP is the same conversation as in the Cleus app, with the same memory and bond level. Stories open in Story Studio at cleus.ai/story, where you can keep writing or export EPUB / PDF / DOCX.

›Tools show up, but every call says “Missing API key”.

The Authorization header isn’t reaching Cleus. It must be exactly “Authorization: Bearer clsk_live_…”. If you use CLEUS_API_KEY, export it in your shell profile and restart the terminal before starting your agent.

›Codex says the url field isn’t supported.

Update Codex (npm i -g @openai/codex@latest). On older versions add experimental_use_rmcp_client = true at the top of ~/.codex/config.toml, or connect through the mcp-remote bridge shown under Claude Desktop.

›Is my key safe?

Your key only goes to api.cleus.ai over HTTPS and Cleus stores just a hash of it. Don’t commit a raw key in a shared .mcp.json — use ${CLEUS_API_KEY} instead. Revoke keys anytime at cleus.ai/developers.

Give your agent Cleus

Create a key, paste one command, and ask for a trailer, a soundtrack, a character or a book.