Introducing Fluence MCP: Deploy and Manage Compute From Your AI Agent

We’ve shipped an MCP server for Fluence, and the path to a running server just got a lot shorter. Connect Cursor, Claude Code, VS Code, Codex, or any other MCP-compatible client, tell your agent what you need, and it handles the rest. Once it’s connected, you don’t even need to open the Fluence Console or learn how it works.

Setup is one config snippet and an API key. The server is free to use; you pay only for the compute you run. Paired with x402 payments, the agent can pay for it on its own, too.

What is MCP?

MCP (Model Context Protocol) is an open standard that lets AI agents use external tools. A server publishes the list of actions it supports, and any compatible client can call them. Without it, connecting an agent to Fluence means pasting API docs into the chat or writing your own wrapper. With it, the agent sees what Fluence can do and calls it directly.

Prompts instead of the Console

The usual route to a server on Fluence runs through the Console: pick a config, click through the setup, come back later to manage it. With MCP, you describe what you need in plain language and your agent does the work. It finds a configuration that fits, deploys it, reports on what’s running, and shuts it down when you’re done.

That’s the full lifecycle, automated. And it happens where you already work: the agent that writes your code can now stand up the server that code runs on, in the same conversation.

Set it up

1. In the Fluence Console, open Settings → MCP and pick your client. Any client that supports MCP will work, and there are ready-made configs for the most popular ones: Cursor, Claude Code, VS Code, Codex, Gemini CLI, Copilot CLI, OpenCode, Cline, Kilo Code, Zed, and Continue.
2. Create an API key. The MCP page links straight to your keys.
3. Paste the config into the file the Console names, restart the client, and check that fluence shows up in its MCP list.

For Claude Code, the config lives in .mcp.json in your project root and reads the key from your shell environment:

{

  "mcpServers": {

    "fluence": {

      "type": "http",

      "url": "https://api.fluence.dev/mcp",

      "headers": {

        "X-API-KEY": "${FLUENCE_API_KEY}"

      }

    }

  }

}

The server is hosted at https://api.fluence.dev/mcp, so there is nothing to install or update. Using a client that isn’t on the list? If it supports remote MCP servers over HTTP with custom headers, point it at the same URL and send your key in the X-API-KEY header.

You stay in control

An agent holding your API key acts on your account and spends your balance. Three things keep that predictable:

• Scope the key. API keys have permissions and an expiry time. Create a separate key for your agent, give it only what it needs, and delete it when you’re done.
• Keep the key out of your repo. The only config that lives in your project, Claude Code’s, references an environment variable instead of the key. Every other config is user-level, outside your repo.
• Ask for the price first. Deployments spend real balance. Have the agent show you the configuration and price before it launches anything.

Pair it with x402: the agent pays, too

Everything above runs on your account and your key. An agent that works on its own can bring its own: MCP covers the deploying, x402 covers the paying. An agent with a wallet sends USDC on Base to Fluence over plain HTTP, and the amount lands on a Fluence balance in the same request. No signup, no email, no Console visit: the paying wallet is the account. The agent then signs in with that wallet, creates its own API key, and plugs the key into MCP.

Put together, that’s an agent that pays for its own compute and deploys it, start to finish, with no human in the loop.

Try it now

MCP joins the API and the Terraform provider as another way to run Fluence without opening the Console.

It’s live now. Open the MCP page in Fluence Console, connect your agent, and tell it what you need.