Claude can request video creation through a connected service such as AKOOL MCP. In this workflow, Claude Code sends an image-to-video task, checks its status, and retrieves the result. AKOOL performs the video generation. Claude provides the chat interface you use to describe, inspect, and manage the task.
If you want to generate AI video from Claude, the important distinction is that Claude is not the rendering model. The AKOOL AI video creation platform supplies the generation tools through its Model Context Protocol connection.
TL;DR
- Connect: Add the AKOOL MCP server to Claude Code using your AKOOL Client ID and Client Secret.
- Prepare: Make sure the approved source image is available to the workflow before submitting a generation.
- Generate: Discover the tools and models first, then submit one
video_image2video_createtask and save its task identifier. - Save: Poll the existing job with
video_image2video_results, then save the finished video instead of submitting the request again.
Table of Contents
- Can Claude generate videos?
- What do you need for Claude Code video generation?
- How do you connect the AKOOL MCP server?
- How does an image-to-video MCP task run?
- Which Claude video prompts should you use?
- How do you troubleshoot AI video automation?
- How do credits and model selection work?
- Frequently asked questions
Can Claude generate videos?
Claude can initiate video generation when it is connected to a rendering service through MCP. In this workflow, Claude Code interprets your request, the Model Context Protocol connects Claude to AKOOL tools, and AKOOL performs the actual image-to-video generation. Claude then checks the asynchronous job and returns the completed result.
| Layer | Role | What it does |
|---|---|---|
| Claude Code | Client and conversational interface | Understands your instruction and invokes tools |
| Model Context Protocol | Connection layer | Exposes AKOOL functions as tools Claude can discover and call |
| AKOOL | Generation service | Runs the image-to-video job and returns its result |
AKOOL's official MCP documentation currently exposes account_models, account_credit, video_image2video_create, and video_image2video_results, among other tools. For a broader look at other integrations, see MCP servers for AI video generation.
What do you need for Claude Code video generation?
You need Claude Code, an AKOOL account with valid API credentials, enough AKOOL credits for the selected generation, and an approved source image that the connected workflow can access. Access to Claude and payment for AKOOL generation are separate.
Prepare four things:
- Claude Code. The workflow in this guide uses Claude Code, not Claude Desktop.
- AKOOL Client ID and Client Secret. AKOOL uses the headers
x-client-idandx-client-secretfor MCP authentication. - Generation credits. Use
account_creditto inspect your AKOOL balance. - An approved source image. Use an image you own or have permission to transform.
For preparing a suitable source asset, see AKOOL's image-to-video workflow.
A clean product photograph is a good first test because it gives you a simple success criterion: does the product remain recognizable while the requested camera motion occurs?
How do you connect the AKOOL MCP server?
Use AKOOL's documented Claude Code route. The server endpoint is https://openapi.akool.com/mcp/v1, authenticated with the x-client-id and x-client-secret headers. Keep both credentials out of prompts, screenshots, repositories, and published documentation.
Install Claude Code if needed:
npm install -g @anthropic-ai/claude-code
Then add the AKOOL MCP server. A credential-safe version of AKOOL's documented command is:
claude mcp add akool-mcp https://openapi.akool.com/mcp/v1 \
--header "x-client-id: $AKOOL_CLIENT_ID" \
--header "x-client-secret: $AKOOL_CLIENT_SECRET"
AKOOL's documentation recommends keeping secrets in environment variables rather than committing them to source control.
Verify the connection:
claude mcp list
You should see an entry for the AKOOL server, for example:
akool-mcp http https://openapi.akool.com/mcp/v1
Use the official AKOOL MCP setup instructions for the current setup command and configuration details. Do not copy credentials into a screenshot.
How does an image-to-video MCP task run?
The safest workflow separates discovery, submission, status checks, and retrieval. A new generation may consume credits. Checking the status of an existing job should reuse its task identifier rather than submitting another generation.
1. Discover the available tools and models
Ask Claude:
List the AKOOL tools available in this session. Identify the image-to-video creation and result tools, and show the required inputs without starting a generation.
You should be looking for video_image2video_create and video_image2video_results. AKOOL also exposes account_models so the agent can inspect models available to the connected account.
2. Prepare the image
Use a stable, rights-cleared source image with a clear subject.
Example:
/product-assets/blue-bottle-approved.png
Do not rely on a local path unless Claude Code and the relevant tool can actually access it.
3. Plan before submitting
Ask Claude:
Use my approved product image as the source. Plan one short clip with a slow push-in and a static product. Show the proposed model, supported settings and any available cost information before submission.
This is a planning turn. It should not start a paid generation.
4. Submit one task
After checking the proposal:
Submit the agreed image-to-video request once. Use a gentle camera move, keep the product still, and return the task identifier.
AKOOL documents video_image2video_create as the creation tool.
Save the returned identifier immediately:
Task ID: [SAVED_TASK_ID]
5. Poll the same task
Ask:
Check the existing task using its saved identifier. Report whether it is queued, processing, complete or failed. Do not create another task.
AKOOL notes that many MCP tools are asynchronous and use result tools to poll for completion.
6. Save the result
Once complete:
Return the finished video link and save location. Compare it with the source image. Propose one motion change for a new attempt, keeping the original output.
A simple lifecycle is:
account_models
↓
video_image2video_create
↓
[SAVED_TASK_ID]
↓
video_image2video_results
↓
queued → processing → complete
↓
[RESULT_URL]
Which Claude video prompts should you use?
Use prompts that make the task state explicit. The main safety distinction is whether the instruction merely inspects the session or creates a new paid generation.
| Prompt | What it does | New generation? |
|---|---|---|
| 1. Discovery | Inspects tools and required inputs | No |
| 2. Planning | Reviews model/settings before submission | No |
| 3. Generation | Submits the image-to-video task | Yes |
| 4. Status | Checks the saved task | No |
| 5. Save and revise | Retrieves output and proposes a revision | No, unless you approve another run |
Prompt 1: Discovery
List the AKOOL tools available in this session. Identify the image-to-video creation and result tools, and show the required inputs without starting a generation.
Prompt 2: Planning
Use my approved product image as the source. Plan one short clip with a slow push-in and a static product. Show the proposed model, supported settings and any available cost information before submission.
Prompt 3: Generation
Submit the agreed image-to-video request once. Use a gentle camera move, keep the product still, and return the task identifier.
Prompt 4: Status
Check the existing task using its saved identifier. Report whether it is queued, processing, complete or failed. Do not create another task.
Prompt 5: Save and revise
Return the finished video link and save location. Compare it with the source image. Propose one motion change for a new attempt, keeping the original output.
These Claude video prompts deliberately separate planning from execution. That is useful for any AI video automation workflow where an accidental second submission can consume additional credits.
How do you troubleshoot AI video automation?
Check the current task and connection state before resubmitting anything. Many failures are authentication, file-access, or asynchronous-status problems rather than reasons to create another generation.
| Symptom | First check | Next action |
|---|---|---|
| AKOOL tools are missing | claude mcp list | Recheck MCP setup and restart Claude Code |
| Authentication fails | Client ID and secret | Verify credentials in AKOOL dashboard |
401 Unauthorized | MCP headers | Replace invalid credentials |
| Source image cannot be used | File accessibility | Move or upload the asset somewhere the workflow can access |
| Job stays queued | Saved task ID | Poll the same job later |
| Result says failed | Existing result response | Read the failure before changing inputs |
| No download available | Job status | Confirm task is actually complete |
| Download link expired | Stored output/state | Retrieve a fresh available result rather than blindly regenerate |
| Two jobs appear | Task history | Confirm whether generation was submitted twice |
AKOOL specifically documents 401 Unauthorized as a reason to verify account credentials.
How do credits and model selection work?
Use the connected account's model and credit tools rather than assuming every AKOOL website model is available through the current MCP task. Model availability belongs to the connected account and tool schema, while generation cost can vary by model and settings.
Use:
account_modelsto discover available models.account_creditto check the current credit balance.- The image-to-video tool's returned settings or quote, when available, to understand the pending task.
AKOOL's normal Image-to-Video interface also notes that model pricing differs.
Do not take one API response that deducted a particular number of credits and publish it as the universal cost for Claude Code video generation. The model, resolution, duration, account, and current pricing can change.
Similarly, AKOOL's main platform offers a broader model catalog, but that does not prove every model is exposed through every MCP workflow.
Generate Your First Video Through Claude Code
The useful pattern is simple: discover → plan → submit once → save the task ID → poll → retrieve.
That makes the MCP video generation workflow easier to debug and reduces the risk of confusing a status check with a second paid generation.
If you want to explore the same generation capabilities directly, open AKOOL's AI video creation platform. For setup details, use the official AKOOL MCP documentation.
