# Coinbase AgentKit

> Coinbase Developer Platform toolkit for giving AI agents a crypto wallet and onchain capabilities — wallet management plus composable actions across protocols, framework-agnostic (LangChain, Vercel AI SDK, and others).

- Readiness score: **81/100** (score.v2)
- Integration difficulty: medium
- Chains: base, ethereum, arbitrum-one, optimism, polygon
- Categories: agents, agent-commerce, account-abstraction, developer-tools, infrastructure
- Last verified: 2026-06-16 by composewith.eth research (coinbase/agentkit and docs.cdp.coinbase.com/agent-kit live-verified via WebFetch/HTTP 2026-06-16) (confidence: high)

## Resources

- docs: https://docs.cdp.coinbase.com/agent-kit/welcome
- sdk: https://github.com/coinbase/agentkit

## Good for

- giving an AI agent an onchain wallet
- agents that transact, pay, or call contracts
- agentic commerce backends
- wiring agent actions to onchain protocols

## Avoid for

- apps with no autonomous-agent component (use a normal wallet SDK)
- non-EVM-first agent stacks
- running agent funds in production without spend limits and guardrails

## Capabilities

### Agent wallets (production-ready)
Best for: provisioning and managing wallets an agent controls, scoped, programmatic onchain access
- Agent-controlled funds require explicit spend limits, allowlists, and key-management guardrails before production.

### Composable onchain actions (production-ready)
Best for: giving agents protocol actions (transfers, swaps, payments), framework-agnostic agent integration

## Composes with

- **x402** (payments_and_commerce; supports) — Agents can programmatically pay for x402-gated APIs and services as part of agentic commerce.
- **circle** (payments_and_commerce; supports) — USDC is the natural settlement asset for agent payments and payouts.
- **erc-8004** (agent_identity_and_trust; supports) — ERC-8004 can provide identity, discovery, and reputation for agents acting onchain.
- **eigencloud** (supporting_protocol; supports) — Verifiable compute (EigenCloud) can attest to agent computations or outputs where trust-minimization matters.

---
Canonical: https://composewith.eth/protocols/agentkit · JSON: https://composewith.eth/protocols/agentkit.json