CornerStone MCP x402 Skill – Tools for Agents

Published as cornerstone-autonomous-agent on npm; source: FinTechTonic/autonomous-agent.

A skill that gives agents a set of x402-paid MCP tools: predict tickers, backtest strategies, link bank accounts, and query agent/borrower scores. Agents using this skill pay with Aptos or EVM via the x402 facilitator. Designed for marketplaces where agents can autonomously download and use the skill; also usable from Cursor, OpenClaw/Moltbot, and headless runners.

Why?

Agents need tools, not a full backend. This skill provides run_prediction, run_backtest, link_bank_account, and score tools (get_agent_reputation_score, get_borrower_score, by-email variants). Agents using it talk to an x402 MCP server, pay with their own Aptos and EVM wallets (verify → settle), and get results after minimal setup—whitelist the agent’s addresses, fund wallets, then the agent can call the tools. The server offers both Aptos and EVM payment options; the skill handles 402 → pay → retry so the agent just calls the tool.

Install

From npm (recommended)

npm install cornerstone-autonomous-agent

Copy the package’s .env.example into your project and set X402_FACILITATOR_URL, HUGGINGFACE_API_KEY, LLM_MODEL, and wallet paths. See Config.

From source

git clone https://github.com/FinTechTonic/autonomous-agent.git && cd autonomous-agent
npm install

OpenClaw / Moltbot: See adapters/openclaw/SKILL.md or use your platform’s skill install (e.g. ClawHub).

Quick Start

Important: Another npm package is named autonomous. To run this package without installing, always use the full name: npx cornerstone-autonomous-agent. After you npm install cornerstone-autonomous-agent, the short binary autonomous in that project runs this CLI.

# Run without installing (use full package name so the other "autonomous" package isn’t used)
npx cornerstone-autonomous-agent setup:aptos
npx cornerstone-autonomous-agent setup
npx cornerstone-autonomous-agent addresses
npx cornerstone-autonomous-agent attest:aptos
npx cornerstone-autonomous-agent "Run a 30-day prediction for AAPL"

# Or install first, then you can use the short command in that project
npm install cornerstone-autonomous-agent
npx autonomous setup:aptos
npx autonomous attest:aptos

# Or install globally for the short command everywhere
npm i -g cornerstone-autonomous-agent
autonomous setup:aptos
autonomous "Run a 30-day prediction for AAPL"

# From source:
node src/setup-aptos.js
node src/run-agent.js "Run a 30-day prediction for AAPL"

CLI: This package registers the autonomous binary (short name) and cornerstone-autonomous-agent (matches package name). Use npx cornerstone-autonomous-agent <command> when you haven’t installed the package; use npx autonomous <command> only inside a project that has cornerstone-autonomous-agent installed, or autonomous <command> after a global install.

Config

Copy .env.example to .env and set:

Commands

Use npx cornerstone-autonomous-agent <command> (or, in a project that has this package installed, npx autonomous <command>). From source: node src/<script>.js or npm run <script>.

Crediting Aptos: Testnet has no programmatic faucet—use Aptos testnet faucet. Devnet: APTOS_FAUCET_NETWORK=devnet npm run credit:aptos. See Canteen – Aptos x402.

MCP Tools

Score definitions: Reputation Score measures an agent’s ability to transact using x402. Borrower score measures real borrower behavior.

All paid tools accept both Aptos and EVM; the server returns 402 with multiple options. Use PREFERRED_PAYMENT_ORDER to prefer one chain/asset.

Supported Networks

x402 Flow

Agent (using this skill) calls MCP tool
  → Server returns 402 + payment requirements (single or array of options: USDC, APT, native ETH)
  → Skill picks one option (by preferredPaymentOrder or first), builds payload (Aptos or EVM)
  → Skill calls facilitator /verify then /settle
  → Skill retries request with payment_payload
  → Server returns result + payment_receipt

Multi-asset: The server may return multiple paymentRequirements (e.g. USDC on Aptos, USDC on EVM, APT on Aptos, native ETH on EVM). The client uses preferredPaymentOrder (e.g. ["aptos:2|usdc", "eip155:84532|native"]) to choose, or the first option. Native ETH: client submits the ETH transfer, then sends the txHash in the payload; facilitator verifies on-chain.

Architecture

autonomous/
├── src/
│   ├── agent/           # ReAct agent, LLM, tool wiring
│   ├── lib/
│   │   ├── mcp/          # MCP client + 402 handle/retry
│   │   ├── aptos/        # Aptos wallet, balance, signPayment
│   │   ├── evm/          # EVM wallet, signPayment (Base)
│   │   └── x402/         # Payment types, verify/settle flow
│   ├── cli.js            # CLI router (autonomous / cornerstone-autonomous-agent)
│   ├── run-agent.js      # Skill demo entrypoint (agent + tools)
│   ├── setup.js          # EVM wallet generation
│   ├── setup-aptos.js    # Aptos wallet generation
│   ├── attest-aptos-wallet.js / attest-evm-wallet.js
│   └── show-agent-addresses.js
├── adapters/             # OpenClaw, OpenAI, Anthropic, local
├── .env.example
└── package.json

Core pieces: lib/mcp — MCP client and 402 retry; lib/aptos / lib/evm — wallets and payment signing; lib/x402 — verify/settle; agent/ — LangChain.js ReAct runner and tools (for agents using the skill).

Tech Stack

• Runtime: Node.js 18+
• Runner: LangChain.js (ReAct), OpenAI-compatible LLM (e.g. Hugging Face) for the demo; agents use the tools via their own runtime
• MCP: Model Context Protocol + x402 payment flow
• Chains: Aptos (viem-style + @aptos-labs/ts-sdk), EVM (viem) for Base Sepolia/Base
• Payments: x402 facilitator (verify/settle), local wallet storage

Security

• Wallets: Stored locally (e.g. ~/.aptos-agent-wallets.json, ~/.evm-wallets.json) for the agent using the skill; private keys not logged or sent except as signed payloads to the facilitator.
• Payments: Only verify/settle go to the facilitator; no custody of funds by the MCP server.
• Whitelist: Addresses used by the agent must be allowlisted at the onboarding flow before paid tools succeed.

Capability + adapters

• Capability: Core (src/) — MCP client, x402 flow, tools for agents. No OpenAI/Claw/Anthropic logic in code.
• Adapters: adapters/ — how each platform or marketplace loads the skill:
  • adapters/openclaw/SKILL.md — OpenClaw / Moltbot
  • adapters/openai/openapi.yaml — Custom GPTs / Assistants
  • adapters/anthropic/tools.json — Claude tools
  • adapters/local/README.md — LM Studio, AutoGen, CrewAI

Deployment order

1. x402 facilitator — Use public (e.g. https://x402-navy.vercel.app/facilitator) for full demo; or run local and set X402_EVM_FACILITATOR_URL to public for link_bank_account.
2. MCP server — x402-enabled.
3. Skill runner (demo) — node src/run-agent.js or PM2 (pm2 start ecosystem.config.cjs --only agent-autonomous from repo root).

References

• Canteen App – Aptos x402 — wallet hydration and crediting
• LangChain.js MCP
• Hugging Face Inference – OpenAI-compatible
• evm-wallet-skill — self-sovereign EVM wallet pattern for agents

License

GPL-2.0-only. Use of this software is also subject to the Responsible AI License (RAIL). See LICENSE.md (GPL-2) and RAIL.