Contenu source (brut)
<p>AI agents need structured tools, clear workflows, and the right context to build production-ready apps on Solana.</p><p>Without these, agents guess at API calls, hallucinate endpoints, hardcode fees that should be dynamic, and miss faster paths like Helius Sender and LaserStream.</p><p>To fix this, we’re releasing a suite of tools for <a href="https://www.helius.dev/agents" rel="noopener noreferrer" target="_blank"><span style="text-decoration: underline">agents building on Helius</span></a>:</p><ol class="list-number"><li value=1>Helius MCP Server</li><li value=2>Helius Skills</li><li value=3>Helius CLI</li><li value=4>Helius Claude Code Plugin</li></ol><p>We’ve also relentlessly optimized our TypeScript and Rust SDKs, as well as our documentation, to make it easier for autonomous agents and AI assistants to recommend and use these tools effectively. </p><p>Here’s what each tool does and how they fit together:</p><h2><strong>The Helius MCP Server: The Tools</strong></h2><p>The Model Context Protocol (MCP) is an open-source standard introduced by Anthropic to enable AI models to securely connect to external data sources, tools, and APIs. It replaces custom integrations with a universal, standardized interface. </p><p>The <a href="https://www.helius.dev/docs/agents/mcp" rel="noopener noreferrer" target="_blank"><span style="text-decoration: underline">Helius MCP Server</span></a> exposes Helius—and Solana—to <em>any</em> MCP client.</p><p>Without MCP, an agent trying to check a wallet balance has to open a shell, construct a <code>curl</code> command, send it, parse the JSON output, and hope it got the endpoint and parameters right. </p><p>Multiply that by every API call in a given workflow, and you get slow, fragile, error-prone interactions that get bogged down by the need to spawn shell instances for every single command.</p><p>With the MCP, none of that happens. </p><p>The agent calls <code>getBalance</code><strong> </strong>as a structured tool call and receives a typed response. No shell. No <code>curl</code>. No output parsing—it just works. </p><p>The same applies to <code>parseTransaction</code>, <code>getAssetsByOwner</code>, <code>createWebhook</code>, and all <a href="https://www.helius.dev/docs/agents/mcp/tools" rel="noopener noreferrer" target="_blank">60+ other tools</a>. Agents can now use Helius directly with the right parameters to get structured data in return. </p><p>The Helius MCP server provides tools for:</p><ul class="list-bullet"><li value=1><strong>Autonomous signup:</strong> create accounts, get API keys, upgrade, etc.</li><li value=2><strong>Live blockchain data:</strong> RPC, DAS API, ZK Compression, and more</li><li value=3><strong>Data streaming:</strong> WebSockets (standard and enhanced), LaserStream gRPC (shred-level latency)</li><li value=4><strong>Transaction sending:</strong> Priority Fee API, SPL transfers, and Sender</li><li value=5><strong>Webhooks</strong>: Create, update, and delete event-driven notifications</li><li value=6><strong>Wallet and history:</strong> balances, transfers, transaction parsing, wallet identity, and perform funding analysis</li><li value=7><strong>Solana knowledge:</strong> SIMDs, Agave and Firedancer source references, Helius blog content, and official Solana docs search</li></ul><p><p>The MCP server works in Claude Desktop, Cursor, VS Code, and Windsurf.</p></p><h3><strong>Quick Installation</strong></h3><p>To add the Helius MCP Server with Claude Code, run:</p><pre><code>claude mcp add helius npx helius-mcp@latest</code></pre><p>For Codex, run:</p><pre><code>codex mcp add helius -- npx helius-mcp@latest</code></pre><p>For Cursor, VS Code, or similar tools, add to your MCP config:</p><pre><code>{
"mcpServers": {
"helius": {
"command": "npx",
"args": ["helius-mcp@latest"]
}
}
}
</code></pre><p>Then set your API key using the <code>setHeliusApiKey</code> tool or an environment variable (e.g., <code>HELIUS_API_KEY</code>), and start shipping. </p><p>For new users, your agent can autonomously create an account for you—it generates a keypair, you fund it, and then call <code>agenticSignup</code> to buy your desired plan starting at $1. </p><h2><strong>Helius Skills: The Brain</strong></h2><p>The MCP gives agents access to 60+ tools. However, access alone isn’t enough—an agent with a tool catalog is akin to a developer with documentation tabs open and zero experience. It can <em>call </em>the tools, but it doesn’t know <em>which</em> tool to reach for, <em>how </em>to combine them for a given task, or <em>what mistakes to avoid</em>. </p><p>Skills are the brain. They’re expert instruction sets—carefully crafted prompts with routing logic, reference files, and rules—that teach an agent how to be an expert Solana developer. </p><p>When a skill is loaded, the agent knows that <code>searchAssets</code><strong> </strong>is better than <code>getProgramAccounts</code><strong> </strong>for NFT lookups, that priority fees should be fetched dynamically via <code>getPriorityFeeEstimates</code>, and that transactions should go through Sender for the highest landing rates. Skills give your agent domain expertise.</p><p>Today, we’re launching four main <a href="https://www.helius.dev/docs/agents/skills" rel="noopener noreferrer" target="_blank"><span style="text-decoration: underline">Helius Skills</span></a>:</p><h3>Build</h3><p>The <a href="https://www.helius.dev/docs/agents/skills/build" rel="noopener noreferrer" target="_blank"><span style="text-decoration: underline">Build skill</span></a> is for core Solana development. It routes to the right tools for DAS, Sender, WebSockets, LaserStream, Webhooks, Enhanced Transactions, wallet analysis, and onboarding. This is the default skill for building anything on Solana with Helius. </p><h3>DFlow</h3><p>The <a href="https://www.helius.dev/docs/agents/skills/dflow" rel="noopener noreferrer" target="_blank"><span style="text-decoration: underline">DFlow skill</span></a> is for building trading apps. It combines DFlow APIs (i.e., spot swaps, prediction markets, real-time streaming, Proof KYC) with Helius infrastructure for transaction submission, fee optimization, and on-chain monitoring. </p><h3>Phantom</h3><p>The <a href="https://www.helius.dev/docs/agents/skills/phantom" rel="noopener noreferrer" target="_blank"><span style="text-decoration: underline">Phantom skill</span></a> is for building frontend apps with the Phantom Connect SDK. It covers wallet connections, token gating, NFT minting, crypto payments, portfolio displays, and secure frontend architecture.</p><h3>SVM</h3><p>The <a href="https://www.helius.dev/docs/agents/skills/svm" rel="noopener noreferrer" target="_blank"><span style="text-decoration: underline">SVM skill</span></a> is so that agents can understand Solana protocol internals. Explore how sBPF bytecode gets executed, Alpenglow design considerations, validator economics, and more by fetching live content from the Helius blog, SIMDs, and Agave/Firedancer source code. No API key required.</p><h3>How Skills Work</h3><p>Skills are model-agnostic. Each one ships with pre-built system prompt variants to work across all platforms, and not just Claude. </p><p>Use <a href="http://openai.developer.md" rel="noopener noreferrer" target="_blank"><span style="text-decoration: underline"><strong>openai.developer.md</strong></span></a><strong> </strong>for the OpenAI API, <a href="http://claude.system.md" rel="noopener noreferrer" target="_blank"><span style="text-decoration: underline"><strong>claude.system.md</strong></span></a><strong> </strong>for the Claude API, or <a href="http://full.md" rel="noopener noreferrer" target="_blank"><span style="text-decoration: underline"><strong>full.md</strong></span></a><strong> </strong>for Cursor Rules, ChatGPT custom instructions, or any other tool. See the <a href="https://www.helius.dev/docs/agents/skills/overview" rel="noopener noreferrer" target="_blank"><span style="
