Artlist MCP · Create a Trailer with Claude or ChatGPT

Quick Summary

• Credit: The trailer, demo, and original method belong to Artlist. Editorial organization, document verification and practical prompts are by JQ VISIONS.
• Artlist MCP connects Claude, ChatGPT, or VS Code to your Artlist account's AI tools. The conversation can list models and voices, initiate generations, track status and return results.
• The precise wording is not that the entire finished trailer was produced exclusively at Claude. The planning and most of the generations happened in conversation, a scene was resolved in the Artlist interface and the editing, color, voice and sound design went through editing software.
• The workflow starts with a brief and a common visual language, moves on to cost-effective storyboarding, final images, shot-by-shot animation, review and finishing. Consistency comes from this sequence, not from a single magical prompt.
• In Claude, the connection is made in Settings, Connectors, Add custom connector, using the server https://mcp.artlist.io/mcp and authorizing the Artlist account.
• In ChatGPT, the current Artlist documentation guides you to the quick connector or to Settings, Plugins, Developer Mode, Browse Plugins and creating the connector with the same URL. The interface nomenclature may evolve.
• MCP uses the plan and the same credit reserve as the account. It does not offer unlimited generations, and a started generation cannot be canceled to recover credits.
• The best configuration maintains human approval on tools that spend credits. The agent must present a plan, model, quantity, duration, resolution and estimated cost, when available, before generating.
• In testing, Nano Banana Pro was used for images and Seedance 2.0 for animation. These are examples from the video, not mandatory choices or guarantees of future availability.
• The method is also used for product campaigns: defining angles and hooks, creating five-shot plans, generating coherent stills and only then animating, narrating and editing.

Credit and short answer: was everything really created inside Claude?

The main video is an Artlist production and follows the creation of an underwater trailer through the Artlist MCP on the Claude. Artlist deserves credit for the concept, images, clips and process demonstrated. This article incorporates the original video, organizes the transcription and compares the workflow with the official documentation available.

The short answer is: almost the entire direction and generation was orchestrated from a conversation, but the end result was not entirely within Claude. The creator planned the narrative, generated storyboards, stills and several clips in the chat. For the angler fish reveal, he used the Artlist interface because he needed a type of reference that the Claude stream didn't offer at the time. Then he took the material to DaVinci Resolve for editing, color, voice and sound design.

This distinction makes the case more useful. MCP reduces context switching and centralizes decisions, but does not eliminate direction, selection, continuity, editing or finishing. The agent becomes a creative console; does not automatically transform a sequence of outputs into a finished movie.

1. What Artlist MCP connects to the conversation

MCP stands for Model Context Protocol. In this case, the Artlist server exposes tools that a compatible client, such as Claude, ChatGPT, or VS Code, can discover and use. The model talks to you; MCP translates authorized orders into operations on the Artlist platform.

According to the official documentation, the connector can work with text to image, text to video, image to image, image to video, text-to-speech, speech-to-speech, voice cloning, model and voice query, Style Kits, generation tracking and music with AI. Specific availability depends on the plan, region, catalog and product condition.

Assets are associated with the account and a dedicated MCP session in Artlist. Small images can appear directly in the conversation; Larger images, videos and audio arrive by phone call. The chat maintains the narrative and Artlist maintains the generation and library.

2. Official video: Artlist MCP for Claude

This short video presents the connector proposal for Claude: describe, iterate and save results in the Artlist account without abandoning the conversation. It's a product view, while the main trailer shows the actual process with successes, failures, and exits to the traditional interface.

The promise of not needing prompt engineering should be read as syntax reduction, not as an absence of direction. A clear brief, consistent benchmarks, cost limits and approval criteria continue to produce better decisions.

3. Before you call: account, credits and permissions

Start with an Artlist account and a compatible client. The MCP page allows you to explore the link, but generations depend on an active plan with AI credits. MCP consumes the same credit reserve as the account, it does not create a separate unlimited package.

The documentation warns that a generation cannot be canceled once it has started and that credits are charged to the order. Therefore, the recommended configuration is to allow frictionless, low-risk operations such as listing templates, checking status, or querying voices, and requiring explicit confirmation to create images, video, audio, or music.

Before approving, ask the agent for a short form: objective, model, number of variations, duration, resolution, references used and estimated cost, if the system can obtain it. If no estimate exists, start with a storyboard or a single test plan.

4. How to connect Artlist MCP to Claude, step by step

1. Open the official Artlist MCP page and choose Connect Claude, or use the direct link indicated in the sources. 2. If the automatic flow does not open, go to Claude Desktop and go to Settings, Connectors or Extensions. 3. Choose Add custom connector. 4. Name the connector Artlist and enter https://mcp.artlist.io/mcp. 5. Confirm the connection, log in to Artlist and accept authorization.

6. Reopen or restart Claude if the tools do not appear. 7. In a new conversation, ask it to list Artlist's tools, image and video models, and available voices. 8. Run a read-only check or generate one test image before authorizing a sequence. 9. Keep confirmation enabled for calls that spend credits.

In video, the creator authorizes most tools, but retains approval for image and video generation. This separation is sensible: it lets the agent research options and monitor tasks, while the financial decision remains human.

5. How to connect Artlist MCP to ChatGPT, step by step

1. Open Artlist's official Connect ChatGPT link. 2. If the quick connector does not appear, follow the current documentation: Settings, Plugins and Developer Mode. 3. Open Browse Plugins and choose the button to add a connector. 4. Enter the URL https://mcp.artlist.io/mcp, create the connection and authenticate the Artlist account. 5. Confirm that Artlist is active in the conversation before requesting a generation.

6. Ask what models, voices and operations your account offers. 7. Define in the first order that no generating tool should be executed without your approval. 8. Start with a single asset, validate where it appears in the Artlist library and only then move on to a batch or sequence.

On Team or Enterprise accounts, an administrator may need to enable custom connectors. Menu names may change as ChatGPT and Artlist update the integration; the MCP page and official Help Center should be the most recent operational reference.

6. Official video: Artlist MCP within ChatGPT

The video below was created to demonstrate the integration of Artlist MCP with ChatGPT. The logic is the same as Claude: idea, model choice, generation, iteration and access to the asset from the conversation, with the results saved in the Artlist account.

The client changes, but the discipline does not: define the project, control permissions, approve costs, identify references and evaluate each output before reusing it.

7. Start with the movie, not the first prompt

The trailer works because there is a simple idea: underwater exploration, monumental scale, growing threat and a final reveal. Before generating, transform the idea into a mini bible with premise, duration, format, audience, emotional arc, palette, light, lens, movement, sound, character, environment and prohibited elements.

The Bible must distinguish global rules from decisions by plan. Global rules include the same wetsuit, dark water, particles, depth and contrast. Each shot defines only composition, action, camera and duration. This separation reduces contradictions and avoids rewriting the entire world with each order.

If you use a prompting skill, treat it as a consistency layer. The video itself shares an external skill for images, but the creator clarifies that he did not use it in this project. The skill is an additional resource, not the origin of the trailer demonstrated.

8. Economical storyboarding before final images

The creator requested twelve storyboard frames with a more economical model. The goal was not to obtain final artwork, but to validate scale, progression, coverage and continuity. The lot cost 360 credits on the account and under the conditions shown in the video. This value is an observation from the test, not a universal table.

Storyboard review found an ocular artifact. This is exactly the function of this phase: to discover flaws in anatomy, direction, geography and narrative when it is still cheap to change. Write down in each frame what works, what fails and what rules should be applied to the final image.

A good AI storyboard includes shot ID, narrative objective, shot type, action, camera movement, estimated duration, sound, and continuity criteria. Without these fields, a beautiful grid may still not look like a movie.

9. Final stills: one master image and one shot at a time

After storyboarding, the creator uses Nano Banana Pro to obtain higher quality stills. First, approve the initial image as an atmosphere reference. Then it asks for one image per scene, instead of generating them all in a final mosaic. This allows you to correct each shot without sacrificing resolution or contaminating the composition of the others.

The master image must fix the elements that cannot fluctuate: diver design, water density, dominant color, quality of light, proportion between human and creature, texture and lens feel. For each new still, declare what it inherits from the master image and what changes.

Save the ID or link for each approved generation. In a long conversation, saying that image can be ambiguous. A name like SHOT_04_DOOR_MASTER_v2 creates a much safer operational reference for the agent and team.

10. Animate stills: camera, action and physical interaction

Before clicking on Animate, describe the shot as a choreography: initial framing, camera movement, subject action, environment movement, main event and final frame. Only then ask the agent to propose the prompt and model.

The door scene shows the most common limit. The diver approaches the mechanism, but does not perform the intended interaction perfectly. Hand-object contact, causality and spatial continuity remain difficult. Break down complex actions, use clear references, and accept that a visually strong version may need editing to hide the error.

The video uses Seedance 2.0 in an eight-second clip in 4K. This is the setup chosen in that test, not a universal recommendation. Compare models with the same still and the same objective before committing credits to an entire sequence.

11. When leaving the chat and using the Artlist interface

In the angler fish reveal, the creator needed to use a specific image reference. The flow available in Claude offered start and end frames, but not the exact control desired. The solution was to open the Artlist library, select the reference, choose Seedance 2.0 and use Auto Prompt.

This is not a methodological flaw. A good system chooses the surface that exposes the necessary control. Use chat for planning, batching, verbal iteration, and tracking. Use the visual interface when you need to pick up a back reference, compare versions, adjust parameters, or browse the library.

The rule of thumb is simple: don't force MCP to hide a capability that the application offers better. Preserve the session and IDs, perform the operation in the interface, and return the result to the project with a note about model, reference, and intent.

12. The trailer is born in the editing: color, voice, music and sound design

The clips were taken to DaVinci Resolve. The creator reduced saturation and contrast, applied a film look, added grain, and softened the sharpness typical of some AI output. The values ​​shown are choices for this material, not universal presets.

The editing decides rhythm and causality. The voice leads to the promise, the sound design connects cuts that were born separately and the music supports the climb. The video estimates it takes around two hours to reach the result presented, but this time depends on experience, number of attempts, duration and finishing requirements.

Reserve one pass just for visual continuity and another for sound. Check movement direction, subject position, light intensity, scale, noise, dialogue, levels, transitions and any distorted frames. The last ten percent editorial is what turns the demo into a trailer.

13. Adapt the same system to product ads and campaigns

In the second demonstration, a product photograph enters Claude and serves as the basis for a social campaign. The agent proposes three angles and hooks, organizes a plan of five shots for each reel, suggests models and generates stills before the animation.

For advertising, add rules that a fictional trailer can do without: product geometry and label, brand colors, approved claim, audience, channel, duration, safe areas, CTA and list of elements that can never be invented. Creativity should happen around the product, not rewrite it.

Start with a single concept and a single hero plan. If the identity survives, move on to variations. If the model changes the logo, packaging, proportions or text, return to the base image or render the actual product in the editor rather than accepting an approximation.

14. Current limits that the article does not hide

Artlist identifies MCP as beta. Catalog search and video-to-video are not currently available through the connector. Some tasks require the Artlist application, and customers may display the tools differently.

Generation is probabilistic. An agent may arrange the request, but does not guarantee anatomy, text, physics, identity, rights, or coherence. Always review real people, brands, properties, advertising claims, sensitive data and customer outputs.

Models, prices, credits, resolutions, durations and menus change. Use this page as your method and the official documentation as your current product status. Before an expensive batch, confirm your account catalog and do a representative test.

15. Operational playbook in 12 steps

1. Connect Artlist to the client and audit tools. 2. Set allowances and budget. 3. Write the creative bible. 4. Structure scenes and shots. 5. Generate cost-effective storyboard. 6. Review narrative and continuity. 7. Approve a master image. 8. Generate final stills one by one. 9. Plan movement before animation. 10. Record each generation. 11. Assemble, treat color, voice and sound. 12. Carry out quality control and rights.

For collaborative work, a simple sheet is enough: Shot ID, status, objective, prompt, model, reference, cost, version, asset link and revision note. This layer avoids repeating generations and allows you to reconstruct decisions when the customer requests a change.

The real gain from MCP is keeping intention, tools and results close. Quality continues to depend on human direction, constraints, selection and finishing.