PrivateCloud.ai & WOWify

Use PrivateCloud from your AI assistant

The MCP server connects a compatible AI assistant to your PrivateCloud account. Describe what you want in ordinary language, and your assistant can discover models, inspect their inputs, check costs, and use the tools to do the work.

What is MCP, and why use it?

MCP stands for Model Context Protocol. It is a standard way for AI applications to discover and call external tools. PrivateCloud's MCP server gives your assistant a documented set of tools backed by the WOWify API.

Instead of manually finding an endpoint, reading its parameter list, and writing a request, you can ask your assistant to find the right model and prepare the job. The assistant can retrieve the current schema and supported settings before it submits anything.

This is useful for exploring the catalog, creating images and videos, editing reference media, and building workflows that combine several services. Developers can use the same connection in compatible agents and applications. The underlying API still handles generation, billing, and results.

What can you do with it?

Task	How the assistant helps
Find the right model	Search the published v2 model catalog and documented v1 endpoint inventory for the capability you need.
Understand the inputs	Retrieve required parameters, supported values, defaults, and reference-media requirements.
Check pricing and balance	Read your available credits and request a v2 quote for the exact model settings.
Create or edit media	Submit supported image, video, and audio jobs using your PrivateCloud credits, then check their status and retrieve results.
Use reference files	Prepare uploads for supported image, video, and audio files. The client must send the actual file bytes before using the resulting reference URL.
Combine services	Create an image, use its result URL as input to a compatible video model, and retrieve the finished video. Each paid step is authorized separately.
Use v1 services	Discover documented chat, character, prompt, compliance, image, voice, video, credit, and usage endpoints available through the MCP inventory.

Availability depends on the current catalog, your account permissions, the selected endpoint, and the capabilities of your AI application. The connector does not provide database administration or realtime voice sessions.

Connect your assistant

You need: a valid PrivateCloud account access token, enough credits for paid operations, and an AI application that supports custom remote MCP servers.

Server URL

https://api.lightspeedcloud.ai/mcp

General setup

Open your AI application's connectors, integrations, plugins, or MCP settings and add a custom remote server named PrivateCloud.
Enter the server URL above. Choose OAuth when authentication options are shown. Use automatic discovery and dynamic client registration where supported.
Complete authorization on the PrivateCloud-hosted page by entering your existing PrivateCloud access token. Your AI application receives a separate connection token.
Enable or select PrivateCloud in the conversation and allow the assistant to use its tools.
Start with: “Check my PrivateCloud balance and show me what tools are available.”

Keep your account access token in the authorization form or your client's secure credential settings. You do not need to paste it into the conversation. The standard WOWify upstream API key is handled by the server.

ChatGPT

In a ChatGPT environment with custom MCP support, open Plugins or connector settings, choose Add custom MCP server, enter the URL, and select OAuth. Create and install the connector, then authorize it with your PrivateCloud access token. In Work, start a conversation and select @PrivateCloud. Labels and availability can vary by ChatGPT environment.

Grok

Open Grok connectors, choose New Connector → Custom, and enter the server URL. Complete OAuth on the PrivateCloud page, then enable the connector for your conversation. If an older connection never asked for your access token, remove and add that connector again to restart discovery and authorization.

Claude, Gemini, Copilot, and other major AI models

You can use PrivateCloud tools from compatible applications powered by ChatGPT / OpenAI GPT models, Claude (Anthropic), Grok (xAI), Gemini (Google), Microsoft Copilot, Llama (Meta), DeepSeek, Qwen (Alibaba), and Mistral. These names identify AI products or model families; they do not guarantee that each provider's own chat website supports this connector.

Use the same URL in clients that support remote Streamable HTTP with JSON responses and compatible OAuth. Support is determined by the application connecting to the server; using a particular AI model by itself does not establish an MCP connection. If your application has no custom remote connector feature, its developer must add an MCP client integration.

Developer authentication options

A client with secure custom-header support can send an existing PrivateCloud account token in X-PrivateCloud-Access-Token. An MCP OAuth connection token can be sent as Authorization: Bearer <connection-token>. Do not use the standard upstream API key as the MCP connection token or put credentials in the server URL.

The OAuth flow uses PKCE S256 and registered redirect URIs. Legacy SSE-only clients, unsupported native callback schemes, and clients without compatible OAuth or custom headers do not connect automatically.

Try these requests

Be specific about your desired output, size, duration, and budget. Your assistant should consult the live catalog instead of guessing a model name or setting.

“Find models for high-quality product photography. Compare their supported sizes and quote the cost of one square image before generating.”

“Create a photorealistic image of a futuristic city at sunset. Show me the selected model, settings, and credit cost, then ask for approval.”

“Use this uploaded photo as a reference. Find an editing model that can change the background while preserving the subject. Tell me the cost first.”

“Find a model that supports a 5-second, 720p video from an image. Use the image we just generated and ask me to approve the video cost separately.”

“Search both the v1 endpoint catalog and v2 model catalog for text-to-speech options. Show the required inputs and available voices.”

“Check the status of the existing job and show its result. Do not create a new generation.”

How generation, uploads, and costs work

Discover: the assistant finds a supported endpoint or model and reads its current input schema.
Prepare: it gathers required settings and uploads reference files if the client can perform the upload. A local filename or chat attachment alone is not a server-accessible URL.
Review: for v2 models, it requests a quote for the exact settings. For v1 paid endpoints, there is no universal quote tool; the authorized maximum credit amount is a spending ceiling, not a price estimate.
Approve: confirm the model, inputs, and spending limit before the assistant submits a paid operation. Your client's tool permissions may also require approval.
Generate: the server submits the operation and returns a job or operation ID. The assistant retains that ID and checks the existing job until a result or error is available.
Continue: download the output, or use its result URL in a compatible next step after reviewing that step's cost.

Paid operations use your account's normal API pricing. Costs vary by model and settings; request a fresh quote when those settings change. Each step in a multi-step workflow is billed separately.

Uploaded reference URLs expire after 24 hours. Upload authorization lasts 15 minutes, and supported files can be up to 300 MiB, subject to the selected model's own input limits. Expiration of a signed URL does not delete the stored file.

If a response is lost: check the original job, operation ID, or PrivateCloud history. Reusing an identical request ID with the same inputs retrieves the existing operation; starting again with a new ID can create another paid job. An uncertain submission should not be purchased again just to recover its response.

MCP tools for developers

Tools	Purpose
`list_models`, `get_model_schema`	Discover v2 models and inspect their inputs.
`get_price`, `get_balance`	Quote v2 settings and read account credits.
`run_model`, `get_prediction`	Submit v2 jobs with a UUID request ID and credit ceiling, then retrieve status and outputs.
`upload_file`	Prepare a signed multipart upload; the client posts the bytes and returned fields to the upload URL.
`list_v1_endpoints`, `get_v1_schema`	Discover the documented, active v1 endpoint inventory.
`read_v1`, `run_v1`, `get_v1_result`	Read v1 data or submit budget-limited paid operations and retrieve their results.

Troubleshooting

The connector does not ask for my access token: Choose OAuth and complete the PrivateCloud authorization page. Remove and recreate an older connector if needed. A client without compatible OAuth needs secure custom-header support or a developer integration.
The assistant cannot see PrivateCloud: Make sure the connector is enabled in this conversation and tool use is allowed. Ask it to list the available PrivateCloud tools. A connected model does not automatically mean your application has loaded the MCP tools.
A model or parameter is missing: Ask the assistant to refresh discovery and search both catalogs. Use the current schema to check valid values. The inventories expose published models and documented active endpoints, rather than every internal API route.
A request is rejected or credits are insufficient: Check your balance, account access, required inputs, and spending ceiling. Content checks and endpoint restrictions still apply; read the returned status message before changing the request.
I get a timeout or a rate-limit error: For a rate limit, follow the returned retry guidance. After a generation submission times out, inspect the original job or history before making another paid request.
The assistant cannot use an uploaded file: Confirm that the client completed the actual upload and is supplying the resulting reference URL. Check the file type, size, URL expiration, and the model's input requirements.

Browse the live model catalog · More PrivateCloud documentation