VisionX MCP Documentation
Everything you need after connecting VisionX.Understand tools, capabilities, workflows, costs and best practices.
- Current version
- v0.9.1
- Transport
- Streamable HTTP
- Available tools
- 34
Overview
VisionX MCP turns your AI assistant into an AI production studio. Once connected, you ask for work in plain language and your assistant plans it, prices it and makes it. Everything below assumes you are already connected.
Not connected yet? Setup takes under two minutes on the Connect your Agent page. This documentation picks up from there.
Quick Start
Three asks to prove the connection works and see what it can do.
Check your workspace
Confirm the link and see your VX balance before you spend anything.
What workspace am I connected to?Generate your first image
Images are the cheapest way to find a look before you commit to video.
Create a cinematic portrait of a futuristic astronaut.Generate your first video
Bring an image you like to life. You approve the price first.
Turn this image into a 10 second cinematic video.Capabilities
What you can ask for, when each one is the right tool, and the tools behind it.
Image generation
- What it does
- 20 image engines including Seedream, GPT Image, Nano Banana and FLUX.2. Up to 4K, multiple variants per ask.
- When to use it
- Key art, product stills, character sheets, storyboard frames, and finding a look before you pay for video.
- Related tools
- generate_image
- list_engines
- estimate_cost
Video generation
- What it does
- 19 video engines including Seedance, Sora, Veo, Kling and Grok. Text to video, image to video and video to video, with native audio on by default.
- When to use it
- Hero shots, social cutdowns, product films and anything that has to move.
- Related tools
- generate_video
- generate_audio
- get_job
Image editing
- What it does
- Change what you already have. Background removal, upscaling, colour grading, frame extraction, and plain-language edits.
- When to use it
- You like the shot but need it restaged, cleaner, bigger or on a different background.
- Related tools
- edit_image
- finish_asset
Characters
- What it does
- Cast holds a brand identity. Create one from a single photo, then reference it by name and every render resolves the same person.
- When to use it
- Any campaign where a face, wardrobe or product has to look identical across shots and engines.
- Related tools
- create_cast_member
- list_cast
- search_mentions
Storytelling
- What it does
- Hand over a brief and the Director plans concept, cast and shots, then renders and cuts the film with an approval at each money gate.
- When to use it
- A brief should come back as a finished film rather than a single clip.
- Related tools
- director_start
- create_sequence
- render_film
Cost & Credits
Thinking is free. Making costs credits, and you always see the price first.
Plan
Free
Estimate
Free
Approve
You decide
Generate
Spends VX
Nothing is charged until you approve it. Before any paid generation VisionX returns a quote with the exact price and the settings it was priced on. Your assistant shows you that figure and waits. Say no and nothing moves.
What is free
Checking your workspace, listing engines, estimating a price, enhancing a prompt, cancelling a job, building and reordering sequences, rendering the finished cut, and the ffmpeg finishing kinds such as upscaling. Generation, editing and audio spend VX.
Best Practices
Five habits that cut cost and raise the hit rate.
Ask for a plan before you generate
Planning and storyboarding cost nothing. Get the shot list agreed before a single credit moves.
Reuse references for consistency
Point at a Cast member or an earlier generation instead of describing the same face again.
Generate images before videos
An image is a fraction of the price. Find the look in stills, then commit to motion.
Estimate before expensive renders
Ask for the price on 4K or long durations. The quote is free and it is the real figure.
Organise assets into sequences
Build the cut as you go. Assembly and rendering cost nothing, so there is no reason to wait.
Example Workflows
Four real runs, end to end. Open one to see the goal, the prompts and what comes back.
Product advertisement
- Goal
- Turn a product brief into a finished ad you can post.
- Flow
- 1. Brief
- 2. Product images
- 3. Video
- 4. Upscale
- 5. Export
- Example prompts
Turn this product brief into a shot list. Do not render anything yet.Make 4 hero images of the bottle on a wet stone surface, golden hour.Animate the strongest one into a 10 second cinematic shot. Price it first.Upscale the final clip to 2K and give me the download link.- Expected output
- One 2K clip plus the stills, priced and approved at each paid step.
Social media campaign
- Goal
- One concept, many cutdowns, the same look across every placement.
- Flow
- 1. Concept
- 2. Key art
- 3. Variants
- 4. Cutdowns
- 5. Export
- Example prompts
Give me three campaign concepts from this brief.Make key art for concept two at 2K.Restage the same shot for 9:16 and 1:1.Make a 6 second version and a 15 second version, then render both.- Expected output
- A set of matching stills and cutdowns in every aspect ratio you asked for.
Short film
- Goal
- A brief becomes a cut film, with you approving each money gate.
- Flow
- 1. Brief
- 2. Concept
- 3. Cast
- 4. Shots
- 5. Render
- 6. Cut
- Example prompts
Produce a 30 second film from this brief, budget 800 VX.How is it going?Approved, go ahead.Render the film and give me the file.- Expected output
- A cut film with sound, and a spend that never exceeded the budget you set.
Character consistency
- Goal
- The same person across every shot, whatever engine renders it.
- Flow
- 1. Photo
- 2. Cast member
- 3. Reference
- 4. Shots
- 5. Check
- Example prompts
Add this photo to my Cast as Aryan.Make four portraits of Aryan in different wardrobe.Now put Aryan in a 10 second warehouse shot.Does the face match across all of them?- Expected output
- One locked identity that holds across stills and video, on any engine.
Troubleshooting
What each refusal means, and the shortest way past it.
VisionX cannot generate
Almost always the workspace is not switched on yet, or the balance is empty. Ask your agent to check the workspace: it reports both in one answer. If the workspace is off, request access and we enable it.
Workspace unavailable
The connection is valid but MCP is not enabled for that workspace. Studio Scale and Enterprise are on by default; other plans are switched on per workspace. Talk to us and we will turn it on.
Credits exhausted
Reads and planning still work with an empty balance, but nothing will render. Top up, or ask for a cheaper shape: lower resolution, shorter duration, fewer variants.
Generation failed
Ask your agent to check the job again. Some failures still deliver, and re-checking recovers them for free. If it is genuinely gone, a retry re-runs it as a fresh paid job.
Permission denied
Your seat decides what the connection can do. A viewer seat is read-only on every call, so generation is refused no matter what was approved at sign-in. Ask an owner to change your role.
Model unavailable
A provider is briefly down or the engine is not on your plan. Ask for the engine list and pick another: most looks are reachable on more than one engine.
FAQ
The questions that come up most often.
How much does generation cost?
It depends on engine, resolution and length, so ask for an estimate rather than guessing. A 10 second 1080p cinematic shot is around 55 VX; four 2K images is around 4 VX. Planning, storyboarding, assembly and rendering the cut are free.
Can I cancel a job?
Yes. Ask your agent to cancel it and the running job stops. Cancelling costs nothing.
How do I reuse characters?
Add someone to your Cast once, then just name them in a prompt. Your agent resolves the name to that locked identity, and every render in the campaign holds it, even when the shot routes to a different engine.
Can I edit generated images?
Yes, and without re-uploading anything. Describe the change in plain language and the edit runs against the image you already made.
Can I remove backgrounds?
Yes. Background removal sits alongside upscaling, colour grading and frame extraction. Ask for it on any image or clip you hold.
How long are downloads available?
A download link is signed and lasts about 15 minutes. The asset itself does not expire, so ask for a fresh link any time.
Put your agent to work
Bring a brief. We will set up the workspace and the budget caps, and you can watch it build.