In this tutorial, we will set up the GoldRush MCP Server with Claude Code to query on-chain data across 100+ blockchains using natural language. No scripts, no API wrappers. Just type what you want in plain English and get structured blockchain data back instantly.
We will also install GoldRush Agent Skills to give Claude Code deeper context on how to use the GoldRush API effectively.
Overview
Here is the problem: querying blockchain data is tedious. You need to find the right API endpoint, write the request, parse the JSON response, handle errors, and repeat for every chain. It adds up fast.
This tutorial eliminates all of that. By the end, you will be able to type something like:
"Show me the token balances for vitalik.eth across Ethereum, Base, Polygon, Arbitrum, and Optimism"
...and get a clean, structured response back. No code written.
To get there, we use two tools working in tandem.
GoldRush (powered by Covalent) provides structured blockchain data across 100+ chains through a unified API. They recently shipped an MCP (Model Context Protocol) server that exposes 27+ blockchain data tools to any MCP-compatible AI agent.
Claude Code is Anthropic's command-line AI agent. You talk to it in plain English, and it handles the actual work: calling APIs, processing data, generating reports.
MCP is the bridge. Think of it as a plugin system for AI agents that lets Claude talk to external tools like GoldRush natively, the same way a browser talks to extensions. With it, GoldRush's entire toolkit becomes available as native commands inside Claude Code.
The result: one prompt, multi-chain data, structured output.
The Workflow
By the end of this tutorial, you will have:
- GoldRush CLI installed and authenticated
- GoldRush MCP Server connected to Claude Code
- GoldRush Agent Skills installed for better query context
- A working setup where you can query any wallet, token, or transaction across 100+ chains using natural language
Prerequisites
- Node.js v18 or higher installed
- npm v5.2 or higher. Ships with Node.js v18 by default and is required for
npxcommands used in this tutorial -
Claude Code installed (
npm install -g @anthropic-ai/claude-code) - A paid Anthropic plan or API credits. Claude Code does not work on a free account. You need either a Claude Pro subscription or prepaid API credits. Check your plan at claude.ai/settings before proceeding
- macOS or Linux recommended. The GoldRush CLI stores your API key in the OS keychain. This works natively on macOS and most Linux distros. On Windows, keychain support depends on your environment. If you hit issues, skip to the Manual MCP Setup method in Step 3 instead
- Basic familiarity with the terminal
Step 1: Get a GoldRush API Key
Head over to goldrush.dev and sign up for a free API key. GoldRush has a generous free tier that is more than enough for development and testing.
Once you sign up, copy your API key and save it somewhere safe. We will need it in the next step.
Step 2: Install GoldRush CLI
The original goldrush-mcp-server GitHub repo was archived in February 2026. The GoldRush CLI is the replacement and has native MCP support built in.
Open your terminal and install the CLI globally:
npm install -g @covalenthq/goldrush-cli
Once installed, authenticate with your API key:
goldrush auth
It will prompt you to enter your GoldRush API key. Paste the key you got from Step 1. You should see something like this:
The CLI validates your key and stores it securely in your OS keychain. No need to pass the key around manually.
Step 3: Register GoldRush MCP Server with Claude Code
Now we will register GoldRush as an MCP server so Claude Code can use its blockchain data tools natively.
Run:
goldrush install
You will see a prompt asking which client to install GoldRush MCP into. Select Claude Code.
Use the arrow keys to navigate to Claude Code and hit Enter.
The GoldRush MCP server is now registered with Claude Code.
Alternative: Manual MCP Setup
If you prefer to skip the CLI and add the MCP server manually, you can run:
claude mcp add goldrush -e GOLDRUSH_API_KEY=<YOUR_API_KEY> -- npx -y @covalenthq/goldrush-mcp-server@latest
Replace <YOUR_API_KEY> with your actual GoldRush API key.
Both methods achieve the same result. The CLI approach is just cleaner.
Step 4: Verify MCP Connection
Let us verify that GoldRush is properly connected to Claude Code.
Run in your terminal:
claude mcp list
You should see GoldRush listed with a Connected status:
If you see ✓ Connected, you are good to go.
Important: The /mcp command only works inside a Claude Code session. If you want to check available tools from within Claude Code, open a session first by typing claude in your terminal, then type /mcp.
Step 5: Install GoldRush Agent Skills (Recommended)
GoldRush also provides Agent Skills that give Claude Code additional context on how to use the GoldRush API effectively. Think of skills as a structured knowledge base that helps the AI agent make better decisions about which endpoints to call and how to format the data.
Run:
npx skills add covalenthq/goldrush-agent-skills
You will see the skills installer clone the repo and show you 4 available skills:
goldrush-cligoldrush-foundational-apigoldrush-streaming-apigoldrush-x402
Select goldrush-foundational-api using the spacebar, then hit Enter.
Next, you will see a list of 45 agents to install the skill into. The installer automatically includes all universal agents (Amp, Antigravity, Cline, Codex, Cursor, etc.) and lets you select additional ones.
Scroll down in the Additional agents list and select Claude Code using the spacebar. Then hit Enter.
For Installation scope, select Global so the skill is available across all your projects:
The installer will show you a Security Risk Assessment for the skill. Review it and select Yes to proceed with the installation.
You will also get a one-time prompt to install the find-skills helper skill. Select Yes for that too. It helps your agent discover and suggest relevant skills automatically.
Claude Code now has both the GoldRush MCP tools and the foundational API skill installed. You are ready to query.
Step 6: Start Querying Blockchain Data
Now for the fun part. Open Claude Code:
claude
You are now inside the Claude Code interactive session. Try these queries:
Query 1: Cross-Chain Portfolio Analysis
Show me the complete token portfolio for vitalik.eth across Ethereum,
Base, Polygon, Arbitrum, and Optimism. Calculate the total USD value
on each chain and give me a summary table.
Claude Code will call the token_balances tool for each chain, aggregate the data, and produce a formatted summary. One prompt, five chains, structured output.
Query 2: Wallet Investigation
Analyze this wallet: 0xd8dA6BF26964aF9D7eEd9e03E53415D37aA96045
1. Show me their top 10 token holdings on Ethereum mainnet
2. Show their recent transactions
3. Check if they hold any NFTs
4. Give me an activity summary
This shows Claude Code chaining multiple MCP tool calls autonomously. It figures out which tools to call and in what order.
Query 3: Token Analysis
For USDC (0xA0b86991c6218b36c1d19D4a2e9Eb0cE3606eB48) on Ethereum:
1. Get the token metadata
2. Show me the approval events and security info
3. Check what chains this token exists on
Query 4: Generate a Report
Generate a markdown report analyzing wallet
0x742d35Cc6634C0532925a3b844Bc454e4438f44e including:
- All token holdings across Ethereum, Polygon, and Base
- Transaction volume overview
- Any NFT holdings
- A risk assessment based on activity patterns
Save it as wallet-report.md
Claude Code will fetch the data, analyze it, and write the report to a file. From natural language prompt to structured report in under a minute.
Available GoldRush MCP Tools
Here is a quick reference of the key tools now available in your Claude Code session:
| Tool | Description |
|---|---|
token_balances |
Get all token holdings for a wallet on any chain |
transactions_for_address |
Full transaction history for a wallet |
nft_balances |
NFT holdings with metadata |
getAllChains |
List all 100+ supported chains |
token_holders |
Get holders of any token |
get_token_prices |
Historical and current token pricing |
get_transactions_for_block |
All transactions in a specific block |
get_approval_events |
Token approval history (security analysis) |
And many more across Balance, Transaction, NFT, Pricing, and Security services. You can see the full list by running /mcp inside Claude Code.
Troubleshooting
"goldrush: command not found"
Make sure you installed the CLI globally with npm install -g @covalenthq/goldrush-cli. If you are using nvm, make sure you are on the right Node version.
MCP server not showing as Connected
Run claude mcp list to check the status. If it is not connected, try removing and re-adding:
claude mcp remove goldrush
claude mcp add goldrush -e GOLDRUSH_API_KEY=<YOUR_KEY> -- npx -y @covalenthq/goldrush-mcp-server@latest
"/mcp not working"
The /mcp command only works inside a Claude Code session. Type claude first to start a session, then type /mcp.
"Claude says GoldRush MCP is not connected"
This happens if you are using Claude.ai (the web chat) instead of Claude Code (the terminal tool). GoldRush MCP is only connected to Claude Code. Open your terminal, type claude, and run your queries there.
What is Next
Now that you have GoldRush + Claude Code set up, here are some ideas to explore:
- Build a CLAUDE.md file for your project with custom instructions on how Claude Code should analyze blockchain data
- Combine with the GoldRush Streaming API to get real-time alerts on wallet activity and new DEX pairs
- Try the x402 integration where AI agents can pay for their own data queries using stablecoins on Base
- Write a compliance report generator that takes a wallet address and produces a structured risk analysis
Conclusion
We just turned Claude Code into a blockchain analyst. With GoldRush MCP and Agent Skills installed, you can query wallets, tokens, transactions, and NFTs across 100+ chains using nothing but natural language.
The setup takes about 5 minutes. The power it unlocks is significant. No more writing API wrappers, parsing JSON responses, or building data pipelines for simple on-chain queries. Just ask Claude Code and get structured data back.
If you found this tutorial helpful, follow me on X for more content at the AI x Web3 intersection.
Top comments (0)