7 Best MCP Servers for AI Image and Video Generation in 2026

Key takeaways

• An MCP server is a connector: add its URL to Claude, ChatGPT or Cursor and sign in, and the assistant can call image and video models directly from a conversation.

• fal reaches more than 1,000 image, video, audio and 3D models and can quote prices before running anything, which suits developers.

• Melius reaches 100+ image, video and audio models on one subscription, with upscaling, background removal and resizing built in; every result lands on a canvas you can open later.

• Runway, Higgsfield, Krea and Black Forest Labs each run hosted servers with OAuth sign-in, so there's no API key to paste in a chat app.

• Video generation over MCP runs in the background, and restarting a job instead of checking it again means paying twice.

How these were picked

Only official servers from the model makers or platforms themselves made the list; community wrappers, archived repositories, demos and anything still in beta were left out. The order weighs how many image and video models each server reaches, whether it's hosted with a simple sign-in, which assistants it's documented for, the editing and finishing tools it exposes, how it bills and how it handles long-running video jobs. Details come from each maker's MCP documentation as of November 2, 2026.

MCP servers for image and video at a glance

1. fal: Best for developers who want 1,000+ models through one connector

fal is an inference platform for generative media, and its hosted MCP server launched in March 2026.

What works: The server reaches fal's whole catalog of more than 1,000 models for image, video, audio, speech and 3D. Tools cover searching models, reading a model's input schema, checking prices, recommending a model, running jobs, uploading files up to 90 MB and managing a job queue. You can connect with a fal API key or an OAuth sign-in, and fal documents setup for Claude, Claude Code, ChatGPT, Codex, Cursor, VS Code and others.

What doesn't: Results arrive as raw files, with no workspace to arrange or revisit them, and choosing among a catalog this size takes some knowledge of the models.

Best for: Developers and technical creators building scripted pipelines.

2. Melius: Best for many models plus editing, saved to a canvas

Melius bills itself as the AI-native operating system for creatives, with AI agents for creative work, and its MCP server lets assistants drive that canvas.

What works: An assistant can lay out nodes for 100+ models, from GPT Image 2.5 and Nano Banana Pro to Veo 3.1, Seedance 2.5 and ElevenLabs voices, and run them singly or in bulk. Topaz upscaling to 4x, image and video background removal, multi-ratio resizing and GIF export sit beside generation. Results stay on a canvas in your account for hand edits later.

What doesn't: MCP uploads take images, video and audio but not PDFs.

Best for: Creative teams that want an assistant to set up multi-model projects.

3. Runway: Best for Gen-4.5 and partner models from chat apps

Runway launched its hosted MCP server in May 2026.

What works: Sign in with your Runway account and call @Runway in a chat to generate with Gen-4.5 plus partner models such as Seedance 2.5, Kling 3.0, Veo 3.1, GPT Image 2 and Nano Banana Pro, depending on your plan. Runway documents it for Claude, ChatGPT, Cursor, Grok and Replit, and there's no API key to manage.

What doesn't: It uses standard credits, and Explore Mode's unlimited relaxed generations do not apply over MCP. Team and Enterprise Claude workspaces need an admin to add the connector.

Best for: Runway subscribers who want to generate without leaving their assistant.

4. Higgsfield: Best for packaged ad and motion workflows

Higgsfield's MCP server connects its video-first platform to Claude, ChatGPT and other agents.

What works: Beyond plain generation, it exposes ready-made workflows: Marketing Studio turns a product link into a video ad, Ad Multiplier makes variations and Genjutsu handles motion transfer, object swaps and restyling. Soul characters trained from photos keep a person consistent across clips.

What doesn't: Every generation bills at standard credit rates, since unlimited plans and free generations apply only on Higgsfield's own site, and audio isn't available through ChatGPT.

Best for: Marketers producing ad variations from chat.

5. Krea: Best for cost quotes before each job

Krea's MCP server reaches its image and video models through a hosted endpoint.

What works: Models are called by ID, from Krea 3 for images to video models such as Veo 3.1, and a preflight call quotes the cost before you commit. Jobs run asynchronously, with a widget that checks on them automatically, and canceled jobs aren't billed. It's documented for Claude Code, Codex and Cursor.

What doesn't: Sign-in methods draw on different balances, OAuth from workspace compute units and tokens from a separate API balance, and only owners and admins can create tokens.

Best for: Users who want a price check before every generation.

6. Replicate: Best for reaching Replicate's public model catalog

Replicate's hosted server, announced in August 2025, opens its HTTP API to assistants.

What works: Anything in Replicate's public catalog, from image and video models to upscalers, can run from a chat, and responses can be trimmed to save context. It's documented for Claude, Codex, Cursor, Gemini CLI and Windsurf, and a local version runs with npx.

What doesn't: You paste an API token rather than signing in, and the tools mirror the raw API, so it suits developers more than creatives.

Best for: Developers already using Replicate.

7. Black Forest Labs: Best for FLUX images and video from the maker

Black Forest Labs, the company behind FLUX, runs its own hosted MCP server.

What works: Image generation defaults to FLUX 3 Image, with FLUX 3 Image Klein as a faster option and the FLUX.2 family still available, and FLUX 3 covers video. Tools generate, edit attached images and video, make variations and browse recent results. Sign-in is OAuth, and you pick which BFL organization to bill.

What doesn't: It covers FLUX models only, and billing runs at BFL's API rates rather than a subscription.

Best for: FLUX users who want first-party access in their assistant.

Recraft also runs an official hosted server for raster and vector images, with region editing, background tools, vectorizing and upscaling, though it doesn't generate video.

How to set up an MCP server for image generation

1. In Claude, open Settings, then Connectors, and add a custom connector with the server's URL. In ChatGPT, look for the maker's plugin; in Cursor, add the URL to mcp.json.

2. Sign in when prompted. Prefer OAuth endpoints in chat apps and keep API-key endpoints for scripts and CI.

3. Ask for something small first, such as one image, to confirm the connector works and see how results come back.

4. If the assistant says the connector has no tools, refresh the tool list from the connector menu, or restart the app.

Tips for generating video over MCP

• Video jobs run in the background. When a wait times out, tell the assistant to check the same job ID again, since starting over is a second paid generation.

• Say exactly how many outputs you want, such as "one version of each shot," because some servers remember a saved batch size.

• Use a cost quote where the server offers one, such as fal's pricing lookup or Krea's preflight.

• Check the transport if a connection fails: Runway accepts only Streamable HTTP, while Replicate's hosted endpoint uses SSE.

• In sandboxed environments, allowlist the server's file host so downloads aren't blocked.

Frequently asked questions

What is an MCP server for image generation?

An MCP server for image generation is a connector that lets an AI assistant such as Claude, ChatGPT or Cursor call image and video models directly. You add the server's URL, sign in and ask for an image or clip in plain language, and the assistant runs the model and returns the result.

How many AI models can an MCP server reach?

It ranges from one model family to more than 1,000 models. fal's server lists over 1,000 image, video, audio and 3D models, Replicate opens its public catalog, Melius reaches 100+ image, video and audio models on one subscription and Black Forest Labs covers its own FLUX models.

Can Claude generate images and videos?

Yes, through an MCP connector. Add a server such as fal, Melius, Runway or Black Forest Labs as a custom connector in Claude, and Claude can generate images and video with those models during a conversation.

Do MCP image generators cost money?

Most servers are free to connect, though Higgsfield requires a paid subscription, and generations bill the same way as the maker's app or API, in credits or per run. Some subscription perks, such as Runway's Explore Mode and Higgsfield's unlimited plans, don't apply over MCP.

Why does video generation over MCP take so long?

Video models often take minutes per clip, so MCP servers run them as background jobs and return a job ID for the assistant to check until the clip is ready.

Who owns images and videos made through an MCP server?

Ownership follows each platform's terms and the underlying model's license. Melius says: "You own the content you create on Melius and can use, publish, and distribute it commercially, subject to the underlying model providers' terms. Melius doesn't claim ownership of your work and doesn't use it to train AI models."