Versely + Claude MCP: Generate Video, Image, Music From Claude (2026)

What if Claude could generate a video without you ever leaving the chat? Not "Claude wrote a Sora prompt that you then paste somewhere else." Not "Claude opened a browser tab and clicked around for forty seconds." Actual generation - a finished MP4 streaming back into the same thread you started in, complete with the music track, the captions, and a scheduled post time. That is what shipped today. Versely is now the first AI creative suite available as a Model Context Protocol server, which means Claude can call the entire Versely toolchain - video, image, music, voice, slideshow, movie maker, scheduler - as native tools inside the conversation.

This is a guide for the AI-native power user who already lives in Claude desktop, has wired up the filesystem and GitHub MCP servers, and is wondering when the rest of the internet is going to catch up. The short answer: creative just did. The longer answer is in the steps below.

Developer workspace with laptop showing code and a Claude chat session

What MCP is, in one paragraph for non-developers

The Model Context Protocol is an open standard Anthropic released on November 25, 2024 to solve a single problem: every time a model wants to use a new tool or read a new data source, somebody has to build a custom connector. With dozens of model providers and thousands of tools, that becomes the "N times M" integration nightmare every API category eventually hits. MCP is the USB-C of AI tool integration - write the server once, and any compliant client (Claude, Cursor, ChatGPT, Gemini, Microsoft Copilot, VS Code) can speak to it. You install MCP servers the way you used to install browser extensions; the model gets new capabilities, you do nothing special to invoke them.

In December 2025, Anthropic donated MCP to the Agentic AI Foundation under the Linux Foundation, a directed fund co-founded with Block and OpenAI. The protocol is no longer "Anthropic's" - it is the industry-standard tool layer, the same way HTTP stopped being Tim Berners-Lee's protocol roughly a year after CERN released it. By early 2026 there are more than 10,000 active public MCP servers and 97 million monthly SDK downloads across all languages. Enterprises are not piloting MCP anymore - Stacklok's 2026 software report shows 41% of surveyed software organizations in limited or broad production with MCP servers. This is no longer the frontier. This is the floor.

The MCP server ecosystem in 2026

For context on where Versely sits in this landscape, here is what is already shipping. The five most universally supported servers - meaning they work in every major host - are the official Filesystem server (local file operations with directory sandboxing), GitHub (PR management, code search, issue triage across repos), Context7 (live library documentation injected into prompts), Playwright (browser automation), and Brave Search (web search grounding). Beyond the core five, the heaviest-used integrations include Slack (read channels, summarize threads, post messages), PostgreSQL (direct database querying), Notion (page reading and editing), Linear (issue management), and the cloud platform servers for AWS, GCP, and Azure.

Notice the pattern: every one of those is a developer or knowledge-work tool. There is no creative production server in that list. There was no "type a prompt and get a video back" MCP server. Until now.

Versely shipping as an MCP server means the same Claude session you use to refactor a backend can now spin up a 30-second product reel, generate a soundtrack for it, write the caption in your brand voice, and schedule it to nine social platforms - without context-switching to a separate web app, without copy-pasting prompts, without losing the thread.

Why Versely shipped MCP first in AI creative

The decision came down to a question we kept getting from power users: "I already pay for Claude, I already live in Claude, why am I leaving Claude to make a video?" The honest answer was that there was no protocol-level way to bring video generation into Claude until MCP matured. With Anthropic's December 2025 launch of programmatic tool calling and tool search in the API - both designed specifically to optimize production-scale MCP deployments - the surface was finally stable enough to ship a creative server that the model could actually drive without prompt-engineering gymnastics.

There is a second reason. Creative tools have historically been monolithic - you live inside Adobe, you live inside Canva, you live inside Versely. That model is fine for design-led users who want a canvas. It is wrong for the agentic-AI user who wants to describe what they need and have it appear. The agentic user wants creative as a capability of their assistant, not as a destination. MCP is the only standard that makes that possible without locking the user into a single chat interface.

If you want the deeper story on what agentic chat looks like from inside Versely's own surface, our agentic chat walkthrough covers the multi-model orchestration that powers the same toolchain.

Step-by-step: install the Versely MCP server

The full install takes about three minutes if you already have Claude desktop. Here is the path.

1. Make sure you are on a recent Claude desktop build. Open the app, click Settings, then the Developer tab. You want to see the "Edit Config" button. If you do not, update Claude desktop to the latest 2026 release.

2. Open the config file. Settings - Developer - Edit Config opens claude_desktop_config.json in your default editor. The file lives at ~/Library/Application Support/Claude/claude_desktop_config.json on macOS and %APPDATA%\Claude\claude_desktop_config.json on Windows. If the file is empty, that is fine - Claude creates it the first time you click Edit Config.

3. Add the Versely server block. Inside the top-level object, add an mcpServers key (or extend it if you already have one). The Versely server entry looks like this:

{
  "mcpServers": {
    "versely": {
      "command": "npx",
      "args": ["-y", "@versely/mcp-server"],
      "env": {
        "VERSELY_API_KEY": "vrs_live_xxx_paste_yours_here"
      }
    }
  }
}

4. Grab your API key. Go to Versely's API settings page, generate a key, and paste it into the VERSELY_API_KEY field above. Treat it like a password - it carries your credit balance and your account permissions.

5. Save and restart Claude. Save the JSON file, fully quit Claude desktop, and reopen it. When Claude restarts, it spawns the Versely MCP server in the background, handshakes with it, and pulls down the tool manifest. You will see a small tools icon in the input bar - click it and you should see Versely's generators listed alongside any other servers you have installed.

6. Test the handshake. Type "List the tools you can call from the Versely MCP server." Claude will enumerate them. If anything went wrong - bad JSON, wrong API key, server failed to spawn - this is where you will catch it. The Claude desktop logs at ~/Library/Logs/Claude/mcp*.log are the first place to look.

That is the entire install. From here on, you are typing prompts.

Glowing futuristic interface representing AI tools handed off between systems

5 real prompts to test inside Claude

These are the prompts we use internally to smoke-test a new MCP install. Copy them straight into Claude after the install.

1. Generate an image. "Use Versely to generate a hero image for a brutalist coffee brand called Kibo. Single-origin Ethiopian beans on a rough concrete slab, harsh top-down light, very high contrast. Use Flux Pro Ultra and give me a 16:9 aspect ratio."

Claude will call Versely's generate_image tool, stream the progress, and inline the finished image into the chat. You can ask "regenerate with softer light" and it will call the tool again with the adjusted prompt - no re-pasting.

2. Generate a short video. "Now generate an 8-second video using that image as the first frame. Slow push-in on the coffee beans, steam rising at the end, VEO 3.1 with Kling V3 as fallback."

This fires generate_video with the previous image's URL passed as the first-frame reference. The model handles the chaining; you do not have to manage asset IDs.

3. Generate a soundtrack. "Generate a 12-second ambient soundtrack for that video. Warm, low BPM, hints of soft brass, no vocals. Use Suno V5."

The generate_music tool returns a finished audio file. If you ask "make it a touch slower and trim the intro," Claude regenerates with parameters adjusted instead of starting from scratch.

4. Clone a voice and write a VO. "Use my saved voice profile to record a 15-second voiceover for the Kibo video. Script: 'Single-origin Ethiopia. Bright, citrus-forward, finishes clean. Kibo Coffee Co. - serious about the bean.'"

Versely's voice cloning tool ships in the MCP server. If you have a voice already trained on your dashboard, Claude can reference it by name and synthesize a fresh take.

5. Build a slideshow series. "Build a three-card carousel slideshow from the Kibo image. Card one is the hero, card two is the brew method, card three is the tasting notes. Schedule it to Instagram, TikTok, and Pinterest for Monday 8am ET."

This chains three tool calls - generate_image for the two missing cards, create_slideshow for the composition, and schedule_slideshow_series for the calendar push. Claude orchestrates the dependency graph; you confirm once.

If any single primitive interests you on its own, the AI video generator, AI image generator, and AI music generator pages document the underlying tools and their model menus.

Side-by-side: Versely app vs Claude via MCP

The product question every power user asks at this point is "should I stay in Claude or go to the Versely app?" The honest framing is that they solve different jobs.

Job to be done	Versely app	Claude via MCP
Quick one-off generation	Browse models, tweak knobs, render	One sentence in the existing thread
Multi-asset campaign with approvals	Drag-and-drop calendar UI is faster	Conversational confirmation flow
Long-running batch (50+ assets)	Background tasks dashboard	Background tasks via MCP, status in thread
Iterating on brand voice	Brand voice manager + memory	Inline in Claude with your existing context
Showing a client the work in progress	Shareable preview links	Screenshot the chat or share preview links
Debugging a failed render	Generation log with model metadata	Ask Claude, it queries the same logs

The pattern that emerges in practice: the app is the home for visual-first work where you want a canvas; Claude via MCP is the home for language-first work where you have an idea in sentences and want it in pixels with minimum ceremony. We see most users settle into both - app for the deep sessions, Claude for the "while I'm in the middle of something else" generations.

For users already running the agentic Versely chat, the Claude-via-MCP path is a parallel surface, not a replacement. The chat inside the app routes across GLM 5.1, Claude Opus 4.7, Gemini 3.1 Pro, and GPT-5.1 with the same tool layer. Claude via MCP routes through one model - whichever one you are talking to - but inherits Claude's 1M context window. If you are operating against a giant corpus, the Claude Opus 4.7 1M context guide explains why that detail matters more than it sounds.

Architecture: how tool calls, live progress, and inline previews actually work

For the curious technical reader, here is what is happening under the hood when you type that prompt.

The MCP server runs as a local process spawned by Claude desktop. It is a stateless JSON-RPC handler that exposes tools, resources, and prompts to the client. When Claude decides to call generate_video, it sends a tools/call request over stdio to the Versely server. The server validates the parameters against the published JSON schema, authenticates against the Versely API with the key from your config, kicks off the generation job, and returns an initial response with a job ID and progress URI.

This is where live progress becomes possible. MCP supports resource subscription, so Claude can subscribe to the progress URI and the server streams updates as the render advances - 12%, 34%, 67%, complete. Those updates flow back into the Claude chat as inline status messages, which is why you see "Rendering frame 240 of 480" instead of a static "working..." spinner. When the job finishes, the server returns the asset URL plus a content type, and Claude renders the result inline - image, video player, audio waveform, depending on the tool.

The whole loop is asynchronous and non-blocking. While the video renders, Claude can continue the conversation, queue more generations, or summarize what is happening. The model is not stuck waiting; it is multitasking. That is why a "three-day campaign" prompt can spawn 18 tool calls in parallel and still feel like a coherent chat. If you have ever wired up the GitHub MCP server and watched Claude open five issues in parallel while drafting a sixth, this is the same pattern - just with Sora 2 and Suno V5 on the other end instead of gh issue create.

The Versely MCP server inherits authentication from the API key you configured. Credits decrement on your Versely account, not Claude's. Generated assets live in your Versely library by default, with optional Cloudflare R2 mirrors if you want them. Privacy and data residency are governed by your Versely account settings, not Anthropic's - same as if you were using the app directly.

Glowing circuit board representing data flowing between systems

Why this is the first move in "AI creative wherever you already work"

The bigger picture, beyond Claude: MCP means that any agent surface can now host the Versely toolchain. The same server we built for Claude works in Cursor (creative for marketing pages while you code them), in ChatGPT (when Anthropic's protocol started shipping in OpenAI products, the field flipped from "competing standard" to "the standard"), in Microsoft Copilot, in VS Code, in any of the agentic browsers like Comet, Atlas, and Claude's own browsing mode. We did not have to write a Cursor plugin, a ChatGPT custom GPT, and a Copilot extension. We wrote one MCP server. That is the entire point of an open protocol.

What that means strategically: the assumption for the next five years should be that the user does not come to the creative tool. The creative tool goes to the user, wherever they happen to be running an agent. The app does not disappear - canvases still matter, deep editing still happens in dedicated tools - but the long tail of "I need a thing right now, in the middle of doing something else" moves into whatever assistant the user already has open. Versely shipping MCP first in AI creative is a bet that this is where the category is going, and that being early matters.

If you are skeptical of the protocol's staying power, consider the trajectory. In one year, MCP went from a single-vendor proposal to industry standard supported by every major model provider, with a Linux Foundation governance structure and 41% enterprise production adoption. There is no scenario in 2027 where "AI tools should talk to each other through a standard protocol" gets walked back. The only question is who ships the right server first in each category.

FAQ

Do I need a paid Claude account to use the Versely MCP server?

No. The free tier of Claude desktop supports MCP servers. You do need a Versely account with credits, since the generation happens on Versely's infrastructure. If you are running heavy workloads, a Claude Pro or Max plan is worth it for the rate limits, but the protocol itself does not gate on subscription level.

Does this work in Claude on the web?

Claude desktop is the primary supported client at launch because it can spawn local MCP server processes natively. Web Claude supports MCP through remote MCP servers (HTTP transport), and the Versely server ships both a stdio mode for desktop and a remote mode for web. The remote setup is a one-click connector in Claude settings rather than a JSON edit.

What about Cursor, ChatGPT, and other clients?

Same server, different client config. Cursor's MCP setup is documented in its own settings; ChatGPT supports MCP through its connectors directory; VS Code wires it through GitHub Copilot Chat. The Versely tools surface identically in all of them. If you switch between clients during a workday, your generations all land in the same Versely library.

Does Claude see my generated images and videos?

Yes - that is the point. When a generation completes, the asset URL is returned to Claude, and Claude can analyze the result (Opus 4.7 is multimodal and can score the image, suggest variations, or compare against a reference). You can opt out of having Claude analyze outputs by adding a tool-result handler in your config, but most users want the feedback loop.

Is this safe? What happens to my prompts and assets?

Prompts pass through Claude's standard data policy (no training on consumer data by default) and through Versely's standard policy (no training, generations belong to you). The MCP server is the messenger - it does not log prompt content beyond what you would log in the Versely app. Your API key is the auth boundary; revoking it kills the integration cleanly.

Person reviewing analytics and content across multiple devices

Closing: the chat is now the studio

The last decade of software taught us to context-switch between fifteen tools to ship one thing. The next decade is going to be the opposite - one surface, many capabilities, summoned by sentence. MCP is the protocol that makes that real, and Versely shipping the first creative server is a bet that creators want their generators to live where they already work, not in another tab.

Install the server, paste your API key, and ask Claude to make you a video about whatever you are working on right now. That is the whole pitch. The future of creative tools is not a destination - it is a capability that arrives whenever you ask for it.

If you want to keep exploring the Versely tool catalog directly, the AI video generator, the AI image generator, and the AI music generator pages each cover the underlying primitives that Claude now has access to. The MCP server exposes all of them, and a few more.

Sources