Float MCP — Your agent's wallet

Your AI agent stops the moment it hits a paywall. Float fixes that.

Give Claude Code, Cursor, or any MCP-compatible agent a USDC wallet. When it needs to generate an image, search the web, pull Twitter data, or call any paid API — it pays and keeps going. No API keys. No interruptions. No babysitting.

npx @usefloat/cli install

That's it. 60 seconds and your agent has a wallet.

The problem

You're building an agent. It's mid-task. It needs to:

• Generate an image
• Search Twitter for real data
• Scrape a website
• Call a paid API

It stops. Either it needs an API key you don't have, a credit card on file somewhere, or an account on a platform you've never heard of.

You break flow. You go set it up. You come back. The context is gone.

Float eliminates that entirely.

What your agent gets

Once installed, your agent has access to:

| Tool | What it does | Cost | |------|-------------|------| | generate_image | AI image generation (GPT Image, Flux, and more) | from $0.04 | | twitter_search_tweets | Real-time tweet search | $0.01/call | | twitter_user_tweets | Any user's timeline | $0.01/call | | twitter_user_info | Profile, followers, verification | $0.005/call | | twitter_article | Long-form X/Twitter article by tweet ID | $0.005/call | | float_call | Call any paid x402 API automatically | pay-per-use | | float_pay | Send USDC to any Solana address | network fee only | | check_balance | Check the agent's wallet balance | free | | get_agents | List all agents and their wallets | free | | set_policy | Set daily budgets, per-tx limits, kill switches | free | | kill_agent / resume_agent | Pause or restart an agent's spending | free |

No API keys for any of these. The wallet pays.

How it works

1. Install Float — npx @usefloat/cli install drops the MCP server into Claude Code
2. Fund your agent — deposit USDC to the agent's Solana wallet from usefloat.xyz
3. Build — your agent calls tools, Float handles payments automatically

Your agent's wallet is isolated. You set the rules — daily spend limits, per-transaction caps, kill switches. Float enforces them. If an agent goes rogue, one call kills it.

Multi-agent support

Running a fleet? Each agent gets its own wallet and policy. One owner controls all of them — fund from a shared treasury, set different budgets per agent, kill any of them instantly.

Owner treasury
├── image-agent     ($5/day budget)
├── research-agent  ($2/day budget)
└── payments-agent  ($50/day budget)

Setup

Install

npx @usefloat/cli install

This registers an agent, provisions a Solana wallet, and configures the MCP server in Claude Code automatically.

Environment variables

Set these if you're configuring manually:

| Variable | Required | Description | |----------|----------|-------------| | FLOAT_OWNER_CONTROL_KEY | Yes | Owner auth key from usefloat.xyz | | FLOAT_OWNER_WALLET | Yes | Your owner wallet address | | FLOAT_API_URL | No | Defaults to https://api.usefloat.xyz |

Security

Float wraps all external API responses (tweets, scraped content, third-party data) with a security boundary before your agent processes them. Content from external sources is flagged as untrusted data — your agent is instructed not to act on instructions found within external content. This protects against prompt injection via third-party APIs.

Spending is controlled by policy — not by the agent itself. The agent requests a payment, Float checks the policy (daily budget, per-tx limit, allowed addresses), then executes or blocks it. The agent never has direct access to the private key.

Links

• Dashboard: usefloat.xyz
• Install CLI: npx @usefloat/cli install
• npm: @usefloat/mcp