Solana Agent Kit

Solana Agent Kit Development Guide

Build AI agents that autonomously execute 60+ Solana blockchain operations using SendAI's open-source toolkit. Compatible with LangChain, Vercel AI SDK, and Claude via MCP.

Overview

The Solana Agent Kit enables any AI model to:

• Deploy and manage tokens (SPL & Token-2022)
• Create and trade NFTs via Metaplex
• Execute DeFi operations (Jupiter, Raydium, Orca, Meteora)
• Stake SOL, bridge tokens, register domains
• Run in interactive or fully autonomous modes

Key Features

Quick Start

Installation

# Core package
npm install solana-agent-kit

# With plugins (recommended)
npm install solana-agent-kit \
  @solana-agent-kit/plugin-token \
  @solana-agent-kit/plugin-nft \
  @solana-agent-kit/plugin-defi \
  @solana-agent-kit/plugin-misc \
  @solana-agent-kit/plugin-blinks

Environment Setup

# .env file
OPENAI_API_KEY=your_openai_api_key
RPC_URL=https://api.mainnet-beta.solana.com  # or devnet
SOLANA_PRIVATE_KEY=your_base58_private_key

# Optional API keys for enhanced features
COINGECKO_API_KEY=your_coingecko_key
HELIUS_API_KEY=your_helius_key

Basic Agent Setup

import {
  SolanaAgentKit,
  createVercelAITools,
  KeypairWallet,
} from "solana-agent-kit";
import { Keypair } from "@solana/web3.js";
import bs58 from "bs58";

// Import plugins
import TokenPlugin from "@solana-agent-kit/plugin-token";
import NFTPlugin from "@solana-agent-kit/plugin-nft";
import DefiPlugin from "@solana-agent-kit/plugin-defi";
import MiscPlugin from "@solana-agent-kit/plugin-misc";
import BlinksPlugin from "@solana-agent-kit/plugin-blinks";

// Create wallet from private key
const privateKey = bs58.decode(process.env.SOLANA_PRIVATE_KEY!);
const keypair = Keypair.fromSecretKey(privateKey);
const wallet = new KeypairWallet(keypair);

// Initialize agent with plugins
const agent = new SolanaAgentKit(
  wallet,
  process.env.RPC_URL!,
  {
    OPENAI_API_KEY: process.env.OPENAI_API_KEY!,
  }
)
  .use(TokenPlugin)
  .use(NFTPlugin)
  .use(DefiPlugin)
  .use(MiscPlugin)
  .use(BlinksPlugin);

// Create tools for AI framework
const tools = createVercelAITools(agent, agent.actions);

Plugins & Actions

Token Plugin (@solana-agent-kit/plugin-token)

// Deploy a new token
const result = await agent.methods.deployToken({
  name: "My Token",
  symbol: "MTK",
  decimals: 9,
  initialSupply: 1000000,
});

// Transfer tokens
await agent.methods.transfer({
  to: "recipient_address",
  amount: 100,
  mint: "token_mint_address", // optional, defaults to SOL
});

// Check balance
const balance = await agent.methods.getBalance({
  tokenAddress: "token_mint_address", // optional
});

NFT Plugin (@solana-agent-kit/plugin-nft)

// Create collection
const collection = await agent.methods.createCollection({
  name: "My Collection",
  symbol: "MYCOL",
  uri: "https://arweave.net/metadata.json",
});

// Mint NFT to collection
const nft = await agent.methods.mintNFT({
  collectionMint: collection.collectionAddress,
  name: "NFT #1",
  uri: "https://arweave.net/nft1.json",
});

DeFi Plugin (@solana-agent-kit/plugin-defi)

// Swap tokens via Jupiter
const swap = await agent.methods.trade({
  outputMint: "target_token_mint",
  inputAmount: 1.0,
  inputMint: "So11111111111111111111111111111111111111112", // SOL
  slippageBps: 50, // 0.5%
});

// Create Raydium CPMM pool
const pool = await agent.methods.createRaydiumCpmm({
  mintA: "token_a_mint",
  mintB: "token_b_mint",
  configId: "config_id",
  mintAAmount: 1000,
  mintBAmount: 1000,
});

Misc Plugin (@solana-agent-kit/plugin-misc)

// Compressed airdrop (cost-efficient)
const airdrop = await agent.methods.sendCompressedAirdrop({
  mintAddress: "token_mint",
  amount: 100,
  recipients: ["addr1", "addr2", "addr3"],
  priorityFeeInLamports: 10000,
});

// Get token price
const price = await agent.methods.getPrice({
  tokenId: "solana", // CoinGecko ID
});

Blinks Plugin (@solana-agent-kit/plugin-blinks)

Execute Solana Actions/Blinks directly:

// Execute a Blink
const result = await agent.methods.executeBlink({
  blinkUrl: "https://example.com/blink",
  params: { /* blink-specific params */ },
});

Integration Patterns

LangChain Integration

import { SolanaAgentKit, createSolanaTools } from "solana-agent-kit";
import { ChatOpenAI } from "@langchain/openai";
import { createReactAgent } from "@langchain/langgraph/prebuilt";
import { MemorySaver } from "@langchain/langgraph";
import { HumanMessage } from "@langchain/core/messages";

async function createLangChainAgent() {
  // Initialize LLM
  const llm = new ChatOpenAI({
    modelName: "gpt-4-turbo-preview",
    temperature: 0.7,
  });

  // Initialize Solana Agent Kit
  const solanaKit = new SolanaAgentKit(
    wallet,
    process.env.RPC_URL!,
    { OPENAI_API_KEY: process.env.OPENAI_API_KEY! }
  )
    .use(TokenPlugin)
    .use(DefiPlugin);

  // Create LangChain tools
  const tools = createSolanaTools(solanaKit);

  // Create agent with memory
  const memory = new MemorySaver();
  const agent = createReactAgent({
    llm,
    tools,
    checkpointSaver: memory,
  });

  return agent;
}

// Run agent
async function chat(agent: any, message: string) {
  const config = { configurable: { thread_id: "solana-agent" } };

  const stream = await agent.stream(
    { messages: [new HumanMessage(message)] },
    config
  );

  for await (const chunk of stream) {
    if ("agent" in chunk) {
      console.log(chunk.agent.messages[0].content);
    }
  }
}

Vercel AI SDK Integration

import { SolanaAgentKit, createVercelAITools } from "solana-agent-kit";
import { openai } from "@ai-sdk/openai";
import { generateText } from "ai";

async function runVercelAgent(prompt: string) {
  const agent = new SolanaAgentKit(wallet, rpcUrl, options)
    .use(TokenPlugin)
    .use(DefiPlugin);

  const tools = createVercelAITools(agent, agent.actions);

  const result = await generateText({
    model: openai("gpt-4-turbo"),
    tools,
    maxSteps: 10,
    prompt,
  });

  return result.text;
}

// Usage
const response = await runVercelAgent(
  "Swap 0.1 SOL for USDC using the best rate"
);

MCP Server for Claude

Install and configure the MCP server for Claude Desktop:

# Install globally
npm install -g solana-mcp

# Or run directly
npx solana-mcp

Add to Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "solana": {
      "command": "npx",
      "args": ["solana-mcp"],
      "env": {
        "RPC_URL": "https://api.mainnet-beta.solana.com",
        "SOLANA_PRIVATE_KEY": "your_base58_private_key",
        "OPENAI_API_KEY": "your_openai_key"
      }
    }
  }
}

Available MCP tools:

• GET_ASSET - Get token/asset info
• DEPLOY_TOKEN - Create new token
• GET_PRICE - Fetch token price
• WALLET_ADDRESS - Get wallet address
• BALANCE - Check balance
• TRANSFER - Send tokens
• MINT_NFT - Create NFT
• TRADE - Execute swap
• REQUEST_FUNDS - Get devnet SOL
• RESOLVE_DOMAIN - Lookup .sol domain
• GET_TPS - Network throughput

Autonomous Mode

Run agent in fully autonomous mode:

import { SolanaAgentKit } from "solana-agent-kit";

const agent = new SolanaAgentKit(wallet, rpcUrl, options)
  .use(TokenPlugin)
  .use(DefiPlugin);

// Configure autonomous behavior
const autonomousConfig = {
  intervalMs: 60000,        // Check every minute
  maxActions: 100,          // Max actions per session
  errorRecovery: true,      // Auto-retry on failures
  dryRun: false,            // Set true for testing
};

// Start autonomous loop
async function runAutonomous() {
  while (true) {
    try {
      // Agent decides what to do based on market conditions
      const decision = await agent.analyze({
        context: "Monitor my portfolio and rebalance if needed",
        constraints: [
          "Keep at least 1 SOL for gas",
          "Max 10% allocation per token",
        ],
      });

      if (decision.shouldAct) {
        await agent.execute(decision.action);
      }

      await sleep(autonomousConfig.intervalMs);
    } catch (error) {
      if (autonomousConfig.errorRecovery) {
        console.error("Error, recovering:", error);
        await sleep(5000);
      } else {
        throw error;
      }
    }
  }
}

Creating Custom Actions

Extend the agent with custom actions:

import { Action, Tool, SolanaAgentKit } from "solana-agent-kit";

// Define the Tool (tells LLM HOW to use it)
const myCustomTool: Tool = {
  name: "my_custom_action",
  description: "Does something custom on Solana",
  parameters: {
    type: "object",
    properties: {
      param1: {
        type: "string",
        description: "First parameter",
      },
      param2: {
        type: "number",
        description: "Second parameter",
      },
    },
    required: ["param1"],
  },
};

// Define the Action (tells agent WHEN and WHY to use it)
const myCustomAction: Action = {
  name: "my_custom_action",
  description: "Use this when you need to do something custom",
  similes: ["custom thing", "special operation"],
  examples: [
    {
      input: "Do the custom thing with value X",
      output: "Custom action executed with param1=X",
    },
  ],
  handler: async (agent: SolanaAgentKit, params: any) => {
    const { param1, param2 } = params;

    // Your custom logic here
    const connection = agent.connection;
    const wallet = agent.wallet;

    // Execute Solana operations...

    return {
      success: true,
      result: `Executed with ${param1}`,
    };
  },
};

// Register custom action
agent.registerAction(myCustomAction);
agent.registerTool(myCustomTool);

Best Practices

Security

1. Never expose private keys - Use environment variables
2. Use dedicated wallets - Separate agent wallet from main funds
3. Set spending limits - Implement max transaction amounts
4. Test on devnet first - Always test before mainnet
5. Audit agent actions - Log all operations

Performance

1. Use appropriate RPC - Helius, Triton for production
2. Batch operations - Combine related transactions
3. Handle rate limits - Implement backoff strategies
4. Cache when possible - Price feeds, token metadata

Agent Design

1. Limit plugin scope - Only load needed plugins (reduces hallucinations)
2. Provide clear context - Detailed prompts improve accuracy
3. Add constraints - Prevent unwanted actions
4. Monitor and iterate - Review agent decisions

Guidelines

• Always test on devnet before mainnet
• Set maximum transaction limits
• Monitor agent activity logs
• Use dedicated wallets for agents
• Implement proper error handling
• Keep private keys secure

Files in This Skill

solana-agent-kit/
├── SKILL.md                           # This file
├── resources/
│   ├── actions-reference.md           # Complete actions list
│   ├── plugins-guide.md               # Plugin deep dive
│   └── security-checklist.md          # Security best practices
├── examples/
│   ├── langchain/                     # LangChain integration
│   ├── vercel-ai/                     # Vercel AI SDK
│   ├── mcp-server/                    # Claude MCP setup
│   └── autonomous-agent/              # Autonomous patterns
├── templates/
│   └── agent-template.ts              # Starter template
└── docs/
    ├── custom-actions.md              # Creating custom actions
    └── troubleshooting.md             # Common issues

Notes

• Solana Agent Kit is actively maintained (1,400+ commits)
• V2 introduced plugin architecture (migration guide available)
• Python version available: solana-agent-kit-py
• MCP server enables Claude Desktop integration
• Apache-2.0 licensed, 1.6k+ GitHub stars