Neurall

MCP

Ask Claude for it and it appears

Neurall is an MCP server. Connect it once and an agent can browse the template library, work out what a generation costs before running it, generate, and bring the finished file back, without you leaving the conversation or writing a line of code.

youHey Claude, create a reveal effect for my logo

  • list_templatesfinds a look worth using
  • get_pricingchecks what it costs first
  • create_generationruns it and waits
Anamorphic Flare, 4s, 1080p with soundone call to video/logo-reveal

A written example. The tool names are the real ones and the clip is a real render, but this is not a recording of a session.

Setting it up, start to first render

Four steps, and the first two are where the money lives. Nothing here needs any code.

  1. 01

    Make a project key

    In the app, open Settings, then Project, then API keys

    A key belongs to one project, never to your account. That scope is the whole safety story here: the agent can only ever reach the one project you hand it, and you can revoke that key without touching anything else.

  2. 02

    Put a balance on that project

    Add, next to the balance in the top bar

    Neurall is prepaid, so the balance is the budget you are handing over. Browsing the template library and asking what a generation costs are free and always will be, which means you can connect the server and let an agent look around before you put a cent in. It can only spend when it generates, and a generation that fails refunds itself.

  3. 03

    Point your client at the server

    Any MCP client that speaks streamable HTTP, Claude included

    mcp server config

    {
      "mcpServers": {
        "neurall": {
          "type": "http",
          "url": "https://api.neurall.io/v3/mcp",
          "headers": { "Authorization": "nl-<projectId>-<key>" }
        }
      }
    }

    Paste the key in as the whole Authorization value. There is no Bearer prefix: the key is the header, and anything in front of it fails the check.

  4. 04

    Ask it for something

    Start with a question, not a generation: this one costs nothing

    youBrowse the Neurall logo reveal templates and show me three you would pick for a fintech brand, with what each costs.

    It browses the library and reads the price list, both free, and comes back with real previews and real numbers. When you like one, tell it to make that reveal from your logo and it runs the generation, waits for it, and hands you the file.

What it can actually do

Every feature in the catalog, plus the things around it an agent needs to work unsupervised: the library it can browse, the prices it can check, and the files it can fetch for itself.

generate_adMake an ad from a product page or a description
generate_youtube_thumbnailMake a thumbnail from a video URL or its subject
generate_self_portraitRestage a person into a new scene, in 4K
generate_product_shotPut a product somewhere it has never been
generate_avatar_videoMake a face speak, in a chosen voice
generate_logo_revealAnimate a logo into a reveal, with sound
generate_object_revealAnimate a product into a reveal
generate_voiceoverSpeak a script in a catalog or cloned voice
create_generationRun any of the features above in one packed call
get_generationCheck a generation and collect its output URLs
list_generationsLook through everything the project has made
delete_generationRemove a result and purge its stored assets
list_templatesBrowse the tested style library, with previews
get_template_promptTake the prompt behind a preview it liked
get_pricingAsk what a generation costs before spending anything
list_voicesBrowse its own voices and the catalog, with previews
get_voiceLook up a voice, or poll one it is cloning
upload_filePull an image or an audio file in from a URL
create_upload_urlSend a file from your own disk, without a key of its own
create_voiceClone a voice from a sample
delete_voiceDelete a cloned voice it no longer needs

An agent you can leave a budget with

It can ask what something costs before it spends anything, the key is scoped to one project rather than the account, and a generation that fails refunds itself. That is the difference between an agent that can generate and an agent you would actually let run.

What people ask it for

The ask, the tools it reached for in order, and what came back.

youHere is our logo. Make me a four second reveal for the top of the launch video, something with heat in it.

  1. upload_filepulls the PNG in and gets a file id
  2. list_templatesbrowses the reveal library, sees Ember Rush
  3. get_pricingchecks the rate per second first
  4. generate_logo_revealsubmits it at four seconds with sound
  5. get_generationpolls it in, a couple of minutes later
Ember Rush, 4s, 1080p with sound

youBefore we commit: what would it cost to shoot fifty product photos and dub the ad into six languages?

  1. get_pricingone call, every feature on the plan

It answers with the arithmetic and spends nothing. An agent that can price a generation before it runs it is an agent you can leave a budget with.

youTake the photo on our product page and give me the same bottle on marble, in daylight and on a beach.

  1. upload_filetakes the image straight off the page
  2. list_templatespicks three scenes out of the library
  3. generate_product_shotthree generations, one per scene
  4. list_generationscollects all three when they land
Three scenes, the product untouched in each

youClone the voice in this recording, then make our mascot say the new tagline with it.

  1. upload_filebrings the audio sample in
  2. create_voicestarts the clone
  3. get_voicewaits for the voice to be ready
  4. generate_avatar_videothe mascot speaks, using the trained voiceId
A trained voice, on a face that never existed

youI just published this video. Get me two thumbnails that are actually about what happens in it.

  1. get_pricingconfirms the rate
  2. generate_youtube_thumbnailpasses the video link, twice
  3. get_generationreturns the two CDN urls
One with a face, one without, both read from the video rather than its title

Worked examples, not recordings: the tool sequences are the ones these asks produce and every result shown is a real render, but the conversations are written rather than captured.

Connect it in a minute

A key, a balance, and one block of config. Browsing the library and checking prices costs nothing, so you can connect it and look around before you put anything in.