Modexia-mcp
The financial infrastructure for autonomous AI. Equips Claude and other agents with secure, programmable USDC smart accounts (ERC-4337). Tools exposed: get_balance, transfer_usdc, get_history. Features: Abstracted gas fees (Paymasters), native USDC settlement on Base, and strict developer-defined spending limits (Policy-as-Code) to eliminate LLM hallucination liability.
Modexia AgentPay: Deep Context & Developer Guide
1. Introduction & Overview
Modexia is a financial infrastructure layer built specifically for autonomous AI agents. It bridges Web2 authentication (Supabase) with Web3 Smart Contract Accounts (ERC-4337 on the Base Network).
This allows an AI agent to programmatically control a non-custodial wallet and make fast, cheap payments (USDC) to other agents, services, or users without needing to manage complex crypto infrastructure directly.
2. Architecture & Security Model
2.1 Current State: API Key & Managed Wallets
Currently, ModexiaAgentpay prioritizes developer experience and simple integration. Agents interact with the protocol primarily through API Keys. The Modexia platform acts as a trusted facilitator that manages the underlying blockchain complexity on behalf of the agent.
Flow:
- Developer Setup: A developer registers on the Modexia platform and receives an API Key. Underlying this account is a Circle Developer-Controlled Wallet (managed by Modexia).
- Channel Opening (On-Chain): When the developer's agent needs to pay a provider, the backend (using the API Key for auth) triggers a transaction to lock USDC into the
ModexiaVaultsmart contract. - Usage (Off-Chain): The agent consumes the provider's API. The Modexia backend tracks the usage and the amount owed off-chain in its database.
- Settlement: When the consumption ends, the Modexia backend (acting on behalf of the agent) submits the final owed amount to the
ModexiaVaultcontract viasettleChannel(). The Vault distributes the USDC to the provider, takes a platform fee, and refunds the rest to the agent's wallet.
2.2 Future State: Trustless State Channels (EIP-712)
Eventually, Modexia will upgrade to Fully Trustless State Channels using cryptographic signatures, where the agent environment manages its own private key and signs "Receipts" (micro-payments) off-chain, which the provider directly submits to the ModexiaVault contract.
3. SDK Capabilities (Python)
Currently, Modexia is interacted with via the Python SDK (modexiaagentpay).
Installation
pip install modexiaagentpay
Initialization
Authentication and environment selection are handled via the API Key prefix:
mx_test_...-> Targets Sandbox (Base-Sepolia)mx_live_...-> Targets Production (Base-Mainnet)
from modexia import create_client
client = create_client(api_key="mx_test_...")
Core Operations
1. Checking Balance
Returns the available USDC balance of the Agent's Smart Contract Wallet.
balance = client.retrieve_balance() # or client.get_balance()
print(f"Balance: {balance} USDC")
2. Direct Payment (Transfer Funds)
Sends USDC to a specific wallet address. The platform charges a 1% fee on top of the amount.
receipt = client.transfer(
recipient="0x7d5...",
amount=5.00, # Amount in USDC (Decimal)
wait=True, # Poll for blockchain finality (Recommended)
idempotency_key="task_123" # Optional: Prevent double-spend
)
3. Smart Fetch (x402 / Paywall Negotiation)
A wrapper around HTTP GET. Automatically detects 402 Payment Required, parses the WWW-Authenticate header for price/destination, pays the invoice via Modexia, and retries the request with a proof-of-payment header.
# Automatically pays if status is 402
response = client.smart_fetch("https://premium-api.com/data")
4. Vault Channels (Micro-payments)
For high-frequency or streaming capability:
client.open_channel(provider, depositAmount)locks funds on-chain.client.consume_channel(channelId, amount)pays a small amount instantly via backend tracking.client.settle_channel(channelId)closes the channel and settles on-chain.
4. Building RAG & AI Agents with Modexia
When building AI Agents using Modexia (especially with TS tools like Vercel AI SDK or MCP), the Agent needs precise context.
RAG Integration
If your agent receives a request to perform a payment, but needs to check policies or external constraints first, it should leverage Retrieval-Augmented Generation (RAG). By reading this documentation (often exposed via MCP resources), the AI can understand that a transfer must be provided a valid 0x... string and a decimal amount.
Best Practices for AI Integration
- Idempotency Keys: Agents should ALWAYS generate a unique standard string (like a combination of the task ID and step) for the
idempotency_keywhen calling payment features to avoid double-charging if the AI retries the action. - Error Handling: AI agents should be aware of
ModexiaPaymentError(insufficient funds, limit exceeded) andModexiaAuthError. If a 402 Error is received dynamically, agents can use thesmart_fetchcommand or manually instruct the transfer command. - Gas Fees: The AI should know that it does NOT need ETH. Gas fees are sponsored by the Modexia Paymaster. Limits are purely enforced in USDC.
Máy chủ liên quan
Scout Monitoring MCP
nhà tài trợPut performance and error data directly in the hands of your AI assistant.
Alpha Vantage MCP Server
nhà tài trợAccess financial market data: realtime & historical stock, ETF, options, forex, crypto, commodities, fundamentals, technical indicators, & more
shadcn/ui
Provides structured data for shadcn/ui components, including descriptions, installation instructions, usage examples, and props.
MCP Playground
A playground for MCP implementations featuring multiple microservices, including news and weather examples.
GhostMCP
Injectable MCP server for AI-driven reverse engineering inside processes
Web Accessibility Testing (A11y MCP)
Test web pages and HTML for accessibility issues and WCAG compliance using Axe-core and Puppeteer.
Authless Remote MCP Server
A remote MCP server without authentication, deployable on Cloudflare Workers or locally with npm.
Apifox
A TypeScript MCP server to access Apifox API data via Stdio.
Claude Code MCP
Orchestrates multiple Claude Code agents across iTerm2 sessions, providing centralized management and inter-agent communication.
Figma (Oficial)
The Figma MCP server brings Figma directly into your workflow by providing important design information and context to AI agents generating code from Figma design files.
Universal MCP Server
A universal MCP server with a modular plugin architecture.
MCP Front
An OAuth 2.1 proxy for MCP servers that enables single sign-on with Google, domain validation, and per-user tokens.