Base L2 Agent Kit
# Base L2 Agent Kit
> MCP server providing Base L2 DeFi tools for AI agents. Wallet management, token swaps, liquidity provision, price feeds, contract interaction, and flash loan arbitrage.
[](https://opensource.org/licenses/MIT)
[](https://base.org)
[](https://modelcontextprotocol.io)
## Overview
Base L2 Agent Kit is an MCP server with **9 tools** for Base L2 operations. Enables AI agents to manage wallets, swap tokens, provide liquidity, read price feeds, and execute flash loan arbitrage — all via typed tool calls.
### Why Base L2?
- **Low gas** — ~$0.01-0.05 per transaction
- **Fast finality** — ~2 seconds
- **Deep liquidity** — Uniswap V3, Aerodrome, Curve
- **Coinbase backed** — Reliable infrastructure
## Tools
| Tool | Description | x402 Price |
|------|-------------|------------|
| `get_wallet_balance` | Check ETH, USDC, any ERC20 | Free |
| `get_transaction_history` | Recent transactions | Free |
| `send_transaction` | Simple transfers | Free |
| `swap_tokens` | DEX aggregation (1inch/Uniswap) | $0.01 |
| `add_liquidity` | Uniswap V3 concentrated positions | $0.03 |
| `remove_liquidity` | Exit + collect fees | $0.03 |
| `get_token_price` | Chainlink + DEX aggregation | $0.001 |
| `call_contract` | Read contract state | $0.001 |
| `execute_flash_loan` | Aave V3 flash loan arbitrage | $0.05 |
## Installation
### Via Smithery (Recommended)
```bash
npx -y @smithery/cli install @ajcheasty2/base-l2-agent-kit
```
### Manual (MCP Client)
```json
{
"mcpServers": {
"base-l2-agent-kit": {
"command": "npx",
"args": ["-y", "@manteclaw/base-l2-agent-kit"],
"env": {
"ALCHEMY_API_KEY": "your-alchemy-key",
"PRIVATE_KEY": "0x..."
}
}
}
}
```
### Docker
```bash
docker run -i --rm \
-e ALCHEMY_API_KEY=your-key \
-e PRIVATE_KEY=0x... \
ghcr.io/manteclaw/base-l2-agent-kit:latest
```
## Quick Start
```typescript
// Check wallet balance
const balance = await client.callTool("get_wallet_balance", {
address: "0x...",
token: "USDC"
});
// Swap ETH → USDC
const swap = await client.callTool("swap_tokens", {
fromToken: "ETH",
toToken: "USDC",
amount: "0.1",
slippage: 0.5
});
// Flash loan arbitrage
const arb = await client.callTool("execute_flash_loan", {
token: "USDC",
amount: "10000",
profitThreshold: 0.5
});
```
## Safety
- 🔒 **Never store keys in plaintext** — Use Bankr, encrypted keystores, or hardware wallets
- 🧪 **All transactions simulated** before execution
- ⛽ **Gas checks** — abort if > $0.10
- 🛡️ **Slippage protection** — revert if exceeded
- 🚫 **No infinite approvals** — exact-amount permits only
- 🧪 **Dry-run mode** available
## Revenue Model
- **x402 micropayments** — $0.001-0.05 per tool call
- **MCP Market** — subscription or per-call pricing
- **Enterprise** — custom integration pricing
## Links
- 🏠 [Repository](https://github.com/manteclaw/litcoiin-solutions)
- 📊 [PulseMCP](https://www.pulsemcp.com/servers/manteclaw-base-agent-kit)
- 🌐 [mcpservers.org](https://mcpservers.org/servers/manteclaw/litcoiin-solutions)
- ⚡ [Smithery](https://smithery.ai/server/@ajcheasty2/base-l2-agent-kit)
## License
MIT © manteclaw
TDQS
Scored across 8 tools
Each tool has a clearly distinct purpose: auditing contracts, getting gas prices, finding yield, analyzing token risks, assessing wallet health, generating mnemonics, fetching prices, and getting DEX quotes. There is no overlap where two tools perform the same function.
Tool names are readable but follow mixed conventions: some use verb_noun (generate_mnemonic), some noun_verb (contract_audit), some noun_noun (wallet_health, dex_quote), and some have a base_ prefix while others don't. The pattern is not consistent enough to be predictable.
With 8 tools, the server is well-scoped within the ideal 3-15 range. Each tool covers a necessary aspect of the Base L2 agent domain without redundancy or bloat.
The set covers a broad range of analytical and informational operations (audit, gas, yield, token analysis, wallet health, price quotes), but lacks actual transaction execution, wallet address derivation, or balance checks. This creates notable gaps for a complete agent workflow, though the existing tools handle a coherent subset.