Can Claude send payments?
Yes. Claude can send payments when connected to MoltPe's MCP server, which exposes payment tools (create agent wallet, check balance, send payment, set spending policy) as first-class MCP tools. Claude Desktop, Claude Code, and any MCP-compatible Claude integration can call these tools autonomously within the spending policies you configure. Your Claude agent never holds raw keys and can never exceed the per-transaction or daily caps you set at the infrastructure level. MoltPe is AI-native payment infrastructure that gives AI agents isolated wallets with programmable spending policies for autonomous USDC stablecoin transactions.
The short version
- MCP connection: Add MoltPe's MCP server to Claude Desktop, Claude Code, or any MCP-compatible Claude host. Setup is a single JSON config entry.
- Tool availability: Claude gets
check_balance,send_payment,list_transactions,call_x402_endpoint, andagent_infoas native tools it can invoke mid-conversation. - Policy enforcement: Per-transaction caps, daily caps, and recipient allowlists are enforced by MoltPe infrastructure — not by Claude's reasoning — so prompt injection cannot bypass them.
In more detail
Claude by itself doesn't move money — a language model has no intrinsic ability to sign blockchain transactions. What gives Claude payment capability is MCP (Model Context Protocol), the standard for exposing tools to an LLM. When you connect MoltPe's MCP server to Claude, the model discovers payment tools the same way it discovers file-read or web-search tools: as structured function calls with schemas it can use on its own initiative.
Once connected, Claude can autonomously check its agent wallet's balance, pay an x402-protected API, call an MPP-metered service, list its recent transactions, or report back what it spent on a task. These happen as MCP tool invocations; MoltPe handles the cryptography, the policy check, and the on-chain transaction. Claude never sees a private key, never signs anything directly, and has no ability to move funds outside the policies you configured.
The security model matters. Because spending policies live at MoltPe's infrastructure layer — below the MCP interface, below Claude's reasoning — a prompt-injected agent cannot exceed its limits. If a malicious user tricks your Claude agent into trying to drain its wallet to an attacker address, the infrastructure simply refuses: the recipient is not on the allowlist, or the amount exceeds the per-transaction cap, or the daily limit is already spent. The signature never gets produced, the transaction never hits the chain.
Related questions
- Does MoltPe work with Claude Desktop?
- Blog: MCP server for AI agent payments
- Blog: Claude Code agent payments setup
- What is MCP (Model Context Protocol)?
- How do AI agents make payments?
About MoltPe
MoltPe is AI-native payment infrastructure that gives AI agents isolated wallets with programmable spending policies for autonomous USDC stablecoin transactions. Live on Polygon PoS, Base, and Tempo.