Revana

Serveur MCP distant pour la génération de vidéos et d’images par IA : Seedance 2.5, Kling, Veo, Nano Banana, GPT Image et bien plus, avec une seule clé API et un paiement à l’usage.

Serveur MCP hébergé

npx add-mcp 'https://revana.si/api/mcp'

S’installe dans Claude Code, Codex, Cursor et plus

Documentation

Make videos from your AI agent

Connect Revana to Claude, ChatGPT, Cursor, VS Code, Codex or any MCP client. Your agent browses every video, image, music and voice model, gets a free quote, makes the job on your prepaid credits once you say yes, and hands you the download link.

Server URL https://revana.si/api/mcp

Paste exactly this, with /api/mcp. The address of this page (revana.si/mcp) is not the server.

Quick start: 30 seconds

  1. Add Revana to your app. One click for Claude, Cursor or VS Code; one command for Claude Code. Every other app is below.
  2. Approve it on revana.si. Sign in (or create an account), check what it can do, press Approve. No key to copy: the app gets its own.
  3. Ask for a video. “Make a 5-second 720p video of a paper boat in the rain. Tell me the price first.” Your agent quotes it, waits for your yes, and gives you the link a few minutes later.

No credits yet? Top up from $10 in your dashboard. Quotes, the model list and checking on a job are always free.

Connect your app

Apps that sign in themselves (Claude, ChatGPT, Cursor, VS Code, Claude Code, Codex) only need the URL: you approve the connection on revana.si and the app gets its own API key, which you can revoke in your dashboard. Apps that take headers can use a key you paste instead.

Claude

claude.ai and Claude Desktop

One click adds Revana as a custom connector. Press Connect, approve it on revana.si, and Revana is in every chat, on the web, desktop and phone.

By hand

  1. Settings, Connectors, Add custom connector.
  2. Name Revana, URL https://revana.si/api/mcp, then Add.
  3. Press Connect and approve on revana.si.
  4. In a chat, open the tools menu and check Revana is switched on.

Claude Code

Claude Code

Add the server, then type /mcp in Claude Code, choose revana and sign in.

claude mcp add --transport http revana https://revana.si/api/mcp

With an API key instead

claude mcp add --transport http revana https://revana.si/api/mcp \
  --header "Authorization: Bearer sr_live_YOUR_KEY"

Add --scope user to have it in every project.

Cursor

Cursor

Opens Cursor with Revana filled in. Cursor asks you to sign in to Revana the first time it uses it.

By hand: ~/.cursor/mcp.json

{
  "mcpServers": {
    "revana": { "url": "https://revana.si/api/mcp" }
  }
}

With a key, add "headers": { "Authorization": "Bearer sr_live_YOUR_KEY" } next to url.

VS Code

VS Code (Copilot agent mode)

Installs the server; VS Code asks you to sign in to Revana when it first starts it. Or run this in a terminal:

code --add-mcp '{"name":"revana","type":"http","url":"https://revana.si/api/mcp"}'

With an API key:.vscode/mcp.json

{
  "inputs": [
    { "type": "promptString", "id": "revana-key", "description": "Revana API key", "password": true }
  ],
  "servers": {
    "revana": {
      "type": "http",
      "url": "https://revana.si/api/mcp",
      "headers": { "Authorization": "Bearer ${input:revana-key}" }
    }
  }
}

ChatGPT

ChatGPT

  1. In ChatGPT, open Plugins (chatgpt.com/plugins), press Add, then choose Add custom MCP server.
  2. Name it Revana and paste https://revana.si/api/mcp as the Server URL.
  3. Keep Authentication on OAuth, tick the box that says you understand, press Create, then approve the connection on revana.si.
  4. No Add custom MCP server option? Turn on Developer mode in ChatGPT under Settings, Apps, Advanced settings.

Codex

Codex CLI

Add the server, then sign in to Revana in your browser.

codex mcp add revana --url https://revana.si/api/mcp
codex mcp login revana

With an API key: ~/.codex/config.toml

[mcp_servers.revana]
url = "https://revana.si/api/mcp"
bearer_token_env_var = "REVANA_API_KEY"

Then export REVANA_API_KEY=sr_live_YOUR_KEY in the shell that runs Codex.

Windsurf

Add Revana to ~/.codeium/windsurf/mcp_config.json (Windsurf Settings, Cascade, MCP servers, View raw config), then refresh the servers.

{
  "mcpServers": {
    "revana": {
      "serverUrl": "https://revana.si/api/mcp",
      "headers": { "Authorization": "Bearer sr_live_YOUR_KEY" }
    }
  }
}

Gemini CLI

Add Revana to ~/.gemini/settings.json, then check it with /mcp.

{
  "mcpServers": {
    "revana": {
      "httpUrl": "https://revana.si/api/mcp",
      "headers": { "Authorization": "Bearer sr_live_YOUR_KEY" }
    }
  }
}

Claude Desktop with a key

The connector above is the easy way. For a key instead (no sign-in), add this to claude_desktop_config.json (Settings, Developer, Edit Config; needs Node.js) and restart Claude.

{
  "mcpServers": {
    "revana": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://revana.si/api/mcp",
        "--header", "Authorization: Bearer sr_live_YOUR_KEY"]
    }
  }
}

n8n

n8n

Give an AI Agent node an MCP Client Tool and set:

  1. Endpoint: https://revana.si/api/mcp
  2. Server Transport: HTTP Streamable
  3. Authentication: Bearer Auth, with your sr_live_ key as the token
  4. Tools to include: All

Any other tool that takes an MCP URL and a header works the same way.

Other MCP clients

Streamable HTTP at one URL. Clients that sign in themselves only need the URL; everything else sends a key:

{
  "mcpServers": {
    "revana": {
      "type": "http",
      "url": "https://revana.si/api/mcp",
      "headers": { "Authorization": "Bearer sr_live_YOUR_KEY" }
    }
  }
}

API keys and sign-in

Sign-in (OAuth). When your app connects, revana.si shows what it may do and whose credits it uses (if you are in a team, you pick the team or your own account). Approve, and the app gets an API key named after it, such as “Claude connector”. It can see models and prices, get quotes, make jobs on your credits and check them. It cannot see your payments or manage your keys. Buying credits stays off unless you tick that box.

A key you paste. Make one in your dashboard under API keys. It looks like sr_live_ plus 64 characters and is sent as Authorization: Bearer sr_live_…. The same key works for the REST API. Keep it private: it spends your credits.

Disconnect. Revoke the app’s key in the dashboard and it stops working at once, in every chat.

Tools

Ten tools. You never call them yourself: ask in plain words and your agent picks them. Every answer ends with a next line telling the agent what to do, so it quotes before spending, waits instead of looping, and explains errors.

ToolWhat it does
list_modelsEvery model, its options and price. Free.
get_quoteThe exact price of a job before it is made. Free.
generate_videoMakes a video, picture, song or voiceover. Spends credits.
get_videoWaits for a job and says how it went. Free.
get_video_linkA download link for a finished job. Free.
upload_fileA link for a local picture, clip or sound. Free.
get_balanceWhat your credits can spend now. Free.
quote_nemesisPrices a Nemesis restage of your clip. Free.
make_nemesisRestages your clip with your own people, products and places. Spends credits.
buy_creditsBuys a credit pack with your saved card, only if you allowed it. Spends money.

list_modelsFree

Every video, image, music and speech model with what it can do (tasks, resolutions, durations, aspect ratios, voices) and its price: per second, per image, per job or per 1,000 characters. Any model with can_generate: true can be made.

ParameterTypeWhat it does
kindvideo, image or audioOnly that kind.
searchstringMatches the name, maker, id or description.

Try: Which video models do 1080p with sound, and what do they cost a second? or Find me the cheapest image model.

get_quoteFree

Prices a job exactly as generate_video would make it, without making it. Answers quoted_usd, an estimate of how long it takes, and up_to_usd when a dearer backup route might be needed. Takes the same parameters as generate_video below, without idempotency_key.

Try: How much is a 10-second 1080p Seedance 2.5 clip with sound? or Quote this on Kling 3.0 and Veo 3.1 Fast and tell me which is cheaper.

generate_videoSpends credits

Makes a video, picture, song or voiceover on your prepaid credits and answers at once with an id. Your agent quotes first and asks you before it calls this. Seedance 2.5 by default; any model from list_models through model.

ParameterTypeWhat it does
modelstringA model id from list_models (seedance-2.5 by default; e.g. seedance-2.5-fast, kling-3.0, veo-3.1-fast, minimax-h3, nano-banana-pro).
promptstring, requiredWhat happens. For speech models (elevenlabs-*), the words to read, up to 5,000 characters.
resolutionstringSeedance 2.5: 480p, 720p (default) or 1080p. Other models: one of the model’s options.resolutions.
durationinteger, secondsSeedance 2.5: 4 to 30, default 5. Other models: within options.durations.
aspect_ratiostring16:9, 9:16, 1:1, 4:3, 3:4, 21:9 or adaptive, as the model lists.
audiobooleanMake sound with the video.
first_framehttps linkA picture to start from (image to video).
end_framehttps linkA picture to end on.
reference_imagesup to 30 linksCharacters, products or a style to follow.
reference_videosup to 10 linksMotion to follow, or the clip to edit. Video input lowers the Seedance 2.5 per-second price. Other models take one.
reference_audioup to 10 linksA voice or music to follow. Other models take one.
modegenerate or editSeedance 2.5: edit changes the first reference video. Other models edit when given one reference video.
max_cost_usdnumber, 0 to 100Optional. The most you will pay: dearer routes are skipped and the charge never goes over it.
idempotency_keystring, requiredA new unique string for this job (a UUID). Sending it again returns the same job and charges nothing more.

Model-specific parameters

ParameterForWhat it does
negative_prompt, quality, seedvariesOn models that take them (not Seedance 2.5).
lyrics, style, instrumentalmusic modelsThe words to sing, the genre and sound, no vocals.
voice, language_codespeech modelsA voice from the model’s options.voices; an ISO 639 code to force the language.
dialoguelist of {text, voice}elevenlabs-v3: up to 50 lines, each in its own voice, read as one take (send prompt as -).
stability, similarity_boost, speednumbersVoice controls (speed 0.7 to 1.2).

Try: Make a 5-second 720p video of a paper boat in the rain. Tell me the price first. or Animate this photo into an 8-second vertical clip with sound.

get_videoFree

Where a job is (queued, running, succeeded or failed), what it cost, and its size, length and sound when done. With wait_seconds it waits on Revana’s side until the job changes, so your agent does not poll in a loop. A failed job says why and what to do next.

ParameterTypeWhat it does
idstring, requiredThe id generate_video or make_nemesis answered with.
wait_seconds0 to 40Wait up to this long for it to finish. 40 is a good value.

Try: Is my video done yet?

get_video_linkFree

A download link for a finished job, good for about an hour. Ask again for a fresh one. Every job is also in your dashboard.

ParameterTypeWhat it does
idstring, requiredThe job’s id.

Try: Give me the download link.

upload_fileFree

Turns a file on your computer into a link to use as first_frame, reference_images, reference_videos, reference_audio, video_url or image_urls. Your agent names the file’s type and size, then PUTs the bytes to upload_url with exactly upload_headers within the hour (agents with a shell, like Claude Code, Cursor and Codex, do this with curl). The link works for 14 days. Files already online can be passed as their https link directly.

ParameterTypeWhat it does
content_typestring, requiredimage/png, image/jpeg, image/webp, video/mp4, audio/mpeg …
bytesinteger, requiredThe exact size in bytes.
filenamestringKept in the link.

Try: Use./product.png as the first frame and make a 6-second 9:16 ad.

get_balanceFree

What the account can spend now, what running jobs are holding (it comes back, less their cost, when they finish), and the total, in dollars and credits.

Try: How many credits do I have left?

quote_nemesisFree

Prices a Nemesis restage: the per-second rate times the clip in plus the video out, which is as long as the clip.

ParameterTypeWhat it does
modelnemesis or nemesis-lite, requiredNemesis: best faces and detail, clips of 4 to 30 s. Lite: cheaper, 2 to 15 s.
clip_secondsnumber, requiredHow long the clip is.
resolution480p, 720p, 1080pDefault 720p.

Try: What would it cost to restage my 12-second clip at 720p?

make_nemesisSpends credits

Restages your clip with your own pictures: the same moves, timing and camera, played by your people, products and places. Only use pictures of people who agreed to it. Usually 5 to 15 minutes.

ParameterTypeWhat it does
modelnemesis or nemesis-lite, requiredAs quoted.
video_urlhttps link, requiredThe clip to restage.
image_urls1 to 8 links, requiredThe pictures that play the clip.
roleslistOne per picture: person, product, location or look. Default: every picture is a person.
target_framebooleanRedraws the first frame with the change made first, so faces and clothes stay true. Default on for Nemesis. Free.
guidepeople, scene, clipLite only: replace the people and keep the background, take everything from the pictures, or follow the clip’s own picture.
soundnone or originalPut the clip’s own sound back (default on Nemesis).
face_meshbooleanLite only: keeps expressions, eyes and lips.
resolutionstringDefault 720p.
promptstringOptional direction.
max_cost_usdnumber, 0 to 100Optional cap.
idempotency_keystring, requiredA new unique string; the same key again returns the same job.

Try: Restage this clip with me in it, using my photo, at 480p.

buy_creditsSpends credits

Charges your saved card for a credit pack, only when you ask for that exact amount. Off until you tick “Allow this connector to buy credits” when you approve the connection (reconnect to turn it on later).

ParameterTypeWhat it does
amount_usd10, 50, 100, 500, 1000 or 5000The pack, in US dollars.
idempotency_key16 to 72 letters, digits, - or _Reused on every retry, so one purchase is never charged twice.

Try: Buy $50 of Revana credits with my saved card.

How a video is made

  1. Quote. get_quote prices exactly what you asked for. Your agent tells you the price.
  2. Your yes. Your agent is told to spend nothing until you agree.
  3. Make. generate_video answers at once with an id; the price is held from your balance.
  4. Wait. get_video with wait_seconds: 40, repeated until it is done. Videos usually take a few minutes.
  5. Download. get_video_link gives a link good for an hour. You pay what it cost; the rest of the hold comes back. A failed job is not charged.

Example prompts

Paste one into a chat with Revana connected.

  • Make a 5-second 720p video of a paper boat sailing down a rainy street at night. Tell me the price first.
  • Turn./hero.jpg into an 8-second 9:16 clip: slow push-in, hair moving in the wind, with sound.
  • Quote the same 10-second 1080p shot on Seedance 2.5, Kling 3.0 and Veo 3.1 Fast, then make the cheapest.
  • Make three 4-second 480p drafts of this prompt with different camera moves, then the best one at 1080p.
  • Edit this clip so it is snowing: https://example.com/clip.mp4
  • Read this script aloud with a calm female voice and give me the MP3.
  • Write a 20-second product ad as three shots, make each one, and give me the links.

Pricing, credits and limits

PriceThe same per-second, per-image or per-character prices as the API, listed on pricing and in list_models. No MCP fee, no subscription.
CreditsPrepaid; 100 credits = $1. Top up from $10 in the dashboard, or with buy_credits (packs of $10 to $5,000) if you allowed it.
QuotesFree. The charge is the quote. If the first route cannot make a job, a backup does and the most it can cost is shown as up_to_usd before you agree; only the route that delivers is charged.
Failed jobsNot charged.
Cap per jobmax_cost_usd up to $100.00; the charge never goes over it.
Call rate20 calls a second per key in bursts. No daily or monthly caps: your credits are the limit.
Waitingget_video waits up to 40 seconds a call.
Filesupload_file links work for 14 days; download links for an hour (ask again for a new one).
RepeatsThe same idempotency_key never makes or charges a second job.

Troubleshooting

“No MCP server found”, or the connection fails right after sign-in

The URL must be https://revana.si/api/mcp, with /api/. revana.si/mcp is this page and revana.si is the home page: sign-in can succeed against them and the app then finds no server. Remove the connector, add it again with the URL above, and approve once more.

It asks me to sign in again, or says the key is not valid (401)

The key was revoked or mistyped. Reconnect in your app (Claude: Settings, Connectors, Revana, Disconnect then Connect), or paste a current key from your dashboard. A key is sr_live_ followed by 64 characters, sent as Authorization: Bearer sr_live_….

The tools do not show up

Check the connector is switched on in the chat’s tools menu (Claude) or picked in the composer (ChatGPT developer mode). In Claude Code and Gemini CLI run /mcp; in Cursor and Windsurf open the MCP settings and refresh. Revana is Streamable HTTP only: a client set to the older SSE transport, or one that opens it with GET, gets 405. Choose HTTP (or Streamable HTTP) in its settings.

Not enough credits (insufficient_credits)

Nothing was charged. Ask your agent for the balance; a lower resolution or a shorter clip costs less. Top up in your dashboard, or, if you allowed it when connecting, ask your agent to buy a pack with buy_credits. Running jobs hold their price until they finish, so the available balance can be lower than the total.

“Add a card to use the free credit” (card_required)

The free credit on a new account works once a card is saved in your dashboard. Saving it does not charge it.

“Generation is being switched on” (generation_not_enabled)

This account can get quotes but not make jobs yet. Write to hello@revana.si and we switch it on.

The video takes a long time, or my agent keeps checking

Videos usually take a few minutes; long or 1080p clips longer, and a Nemesis restage 5 to 15 minutes. The job runs on Revana whether or not the chat stays open. Your agent should call get_video with wait_seconds: 40, which waits on our side instead of polling. If the chat ended, ask in a new one: “Check my last Revana video”, or open it in your dashboard.

Too many requests (429)

Each key may make 20 calls a second in bursts. Wait the seconds the answer names and send the same request again, with the same idempotency_key.

The job failed

get_video says why (error_code, reason) and what to do. A failed job is not charged. content_policy means the model refused the prompt or a picture under its content rules, so change them rather than retrying. model_declined means the model answered in words: rewrite the prompt as a plain description of what to make.

My agent cannot read a file on my computer

Pictures, clips and sounds go in as public https links. Agents with a shell (Claude Code, Cursor, Codex, VS Code) use upload_file and curl. In claude.ai or ChatGPT, put the file somewhere public and paste the link.

Will retrying charge me twice?

No. Every paid call carries an idempotency_key; sending the same key again returns the same job, or the same purchase, and charges nothing more.

Still stuck?

Email hello@revana.si with the app you use, the time, and the request_id or reference from the error. Developers can test the server directly:

curl -s https://revana.si/api/mcp \
  -H "Authorization: Bearer sr_live_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Privacy and security

  • The connection reaches only your own Revana account: its balance, its jobs and the files it uploads. It cannot see your payments or your other keys.
  • Your prompts, pictures and videos are used to make your job and are never shared or used for training. Read the privacy policy and the terms.
  • Every connection has its own key. Revoke it in the dashboard and it stops at once.
  • Spending is prepaid and capped: no job costs more than its quote’s up_to_usd or your max_cost_usd, and nothing tops up your card unless you allowed it and asked.

Building your own integration? The full REST API is in the docs, and the server is listed for MCP registries in server.json.