What is Higgsfield MCP?
On 30 April 2026, Higgsfield (a platform for AI image and video generation) launched its MCP server. To understand what that changes, you first need to understand what MCP is.
MCP stands for "Model Context Protocol". It is an open standard created by Anthropic in 2024, which allows any AI agent (such as Claude, OpenClaw, Cursor) to connect to external tools through a single link. Put simply: it is the equivalent of the USB-C port, but for AI agents. One socket, and every tool plugs into it.
In practice, Higgsfield MCP lets your AI agent access more than 30 visual generation models directly without having to go through the Higgsfield web interface. You describe what you want, the agent picks the right model, runs the generation, and returns the result to you in the conversation.
Before this launch, to generate an image with Higgsfield you had to open their website or app, manually select the model, write your prompt, run the generation, wait, download, then come back to Claude. With MCP, everything happens in the conversation, and the agent can chain several generations in the same session without interruption.
Which models are available?
MCP gives access to the entire Higgsfield catalogue, meaning more than 30 models. Here are the main ones, sorted by use:
| Category | Available models | Typical use |
|---|---|---|
| Image | GPT Image 2, Nano Banana Pro, Soul V2 (and all the other available models) | Portraits, product photos, illustrations, design, UGC |
| Video | Seedance 2.0, Kling 3.0, HappyHorse, WAN 2.7 | Clips of 5 to 15 seconds, up to 4K, all styles |
| Audio | Built-in translation, voice-over, sound design | Soundtracks, dubbing, sound effects for videos |
The agent automatically picks the best model based on what you ask it for, depending on style, resolution and content type. You can of course force a specific model in your prompt if you have a preference (for example "use Seedance 2.0 for this video").
How do you install it in Claude?
Installation literally takes 30 seconds. Here is the full procedure, step by step:
- Open Claude (on the web at claude.ai, in Cowork, or in Claude Code).
- Go to settings: click on your profile, then "Settings".
- Click on "Connectors" in the left-hand menu.
- Add a custom connector by clicking on "Add custom connector".
- Name it "Higgsfield" and paste the URL:
https://mcp.higgsfield.ai/mcp - Click on "Add", then on "Connect".
- Sign in with your Higgsfield account in the window that opens.
- That's it. Ask Claude to generate an image, and it will automatically use Higgsfield.
Once connected, you never need an API key again. Authentication happens once via your Higgsfield account, and all future generations automatically draw on your monthly credit quota. No more juggling API keys to regenerate or secure.
And in OpenClaw?
OpenClaw is an open-source AI agent (the equivalent of Claude Desktop, but self-hosted). It supports MCP natively, but installation is slightly different: instead of going into graphical settings, you give the agent a direct instruction.
Launch OpenClaw and send it this message:
Connect the Higgsfield MCP at https://mcp.higgsfield.ai/mcp. It supports OAuth device flow, discovery at https://mcp.higgsfield.ai/.well-known/oauth-protected-resource. Run the device flow and send me the link like: "Authorize here: https://higgsfield.ai/device?code=XXXXXXXX" Then poll the token endpoint until I authorize and confirm once it succeeds.
The agent will reply with an authorisation link. Click on it, sign in to your Higgsfield account, approve access, and OpenClaw will confirm as soon as authentication succeeds. From there, you can generate images and videos directly from OpenClaw.
The same principle works for Hermes Agent, NemoClaw, Cursor, and any other MCP-compatible client. The URL stays the same: https://mcp.higgsfield.ai/mcp.
How much does it really cost?
The MCP itself is completely free. You pay nothing for the connection or for using the server. On the other hand, each generation consumes credits on your Higgsfield account, exactly as it does from the website. Here are the four plans available:
| Plan | Monthly price | Credits | Who is it for? |
|---|---|---|---|
| Starter | $15 | 200 / month | Exploration, first tests |
| Plus | $49 | 1,000 / month | Regular creators |
| Ultra (the most popular) | $129 to $310 | 3,000 to 9,000 / month | Serious production, solo agencies |
| Business | $71 per seat | 3,000 / month (shared pool) | Agencies, teams of 2 to 15 people |
In practice, what does that look like in terms of actual generations? Here is what the main models cost in credits:
| Generation | Credit cost | How many with the Plus plan (1000 credits)? |
|---|---|---|
| Standard image (Soul V2) | ~0.5 credit | ~2,000 images |
| GPT Image 2 High (4K) | 12 credits | ~83 images |
| Seedance 2.0 video (1080p, 15 seconds) | 135 credits | ~7 videos |