Sign onchain with 1Claw AgentKit on Base

@1claw/agentkit is a hardened MCP server built on Coinbase AgentKit. Your agent gets the full onchain toolkit for Base, but signing happens in a TEE via the Intents API and secrets are resolved from your vault at boot, never touching disk.

1. 1

   Path A: MCP server setup. Clone the repo and run the setup wizard. It asks for your human 1ck_ key and provisions a vault, an agent with Intents API and Base guardrails, a Base signing key, and a read policy on agentkit/*.

   bash Copy

   git clone https://github.com/1clawAI/1claw-agentkit.git
   cd 1claw-agentkit
   npm install
   npm run setup

2. 2

   The wizard prints an MCP config JSON. Paste the JSON the wizard prints into your AI client config. It pairs 1claw-agentkit with the 1claw MCP server under the agent key the wizard provisioned.

   text Copy

   # The wizard outputs something like:
   # {
   #   "mcpServers": {
   #     "1claw-agentkit": { ... },
   #     "1claw": { ... }
   #   }
   # }
   # Paste that output into your client config file.

3. 3

   Set the chain ID for Base Sepolia (testnet) before moving to mainnet.

   bash Copy

   export ONECLAW_CHAIN_ID=84532

4. 4

   Path B: Library API. Resolve secrets and build a Base wallet provider, then send a TEE-signed transaction. Use the agent credentials from the npm run setup wizard. Value is in ETH (0.001 ETH ≈ a few dollars).

   typescript Copy

   import { createBaseMainnetProvider } from "@1claw/agentkit";
   
   const wallet = createBaseMainnetProvider({
     agentApiKey: process.env.ONECLAW_AGENT_API_KEY!,
     agentId: process.env.ONECLAW_AGENT_ID!,
   });
   
   const result = await wallet.sendTransaction({
     to: "0xRecipient",
     value: "0.001",
   });
   
   console.log(`TX: ${result.txHash} (${result.status})`);

5. 5

   Path C: Full-stack starter (Scaffold-Agent). To bootstrap a full-stack onchain dApp instead, use Scaffold-Agent, which wires HSM-backed secrets and Intents signing into a Scaffold-ETH 2 scaffold.

   text Copy

   # See scaffoldagent.xyz and github.com/1clawAI/scaffoldagent_xyz

Your agent now signs and broadcasts transactions on Base through the Intents API, with keys held in the vault TEE and guardrails capping what a prompt injection could ever do.

What this gives you, and what it does not. This is the point where an agent stops reading and starts moving value, and the failure mode changes shape.

• Onchain actions are irreversible. Every other mistake in this course can be rotated, revoked or re-issued; a broadcast transaction cannot.
• Testnets are the whole point of the setup step. An agent that has never been exercised on a testnet is being debugged with real funds.
• Guardrails are the safety net, not the framework. Recipient allowlists, per-transaction caps and daily limits live on the agent and apply regardless of which framework calls the API.