MCP for blockchain data: every Alchemy API in your AI assistant

Open Claude and ask a question you would normally write a script for: "What are the top USDC lending markets right now, and which one has the highest supply APY?"

A few seconds later it answers. Live rates across Aave, Compound, and Morpho, sorted, with the chain each one is on. No script. No copying an endpoint out of the docs. No hunting for the right contract address. You asked, and your assistant read the chain and told you.

That works because your assistant is connected to the Alchemy MCP server: a hosted service that exposes every Alchemy blockchain API as a tool your AI assistant can call in plain conversation. Onchain questions get answered the same way your assistant already answers questions about your codebase.

What just happened

Coding assistants are good at reasoning and bad at guessing. Ask one for live blockchain data without giving it a real way to fetch that data, and it will often invent an endpoint that does not exist or hand you stale numbers from training data. The intelligence is there. The connection to reality is not.

MCP closes that gap. When your assistant is connected to Alchemy over MCP, "what are the top USDC lending markets" stops being a question it pattern-matches and becomes a question it answers by actually reading the chain: calling the contracts, pulling current rates, and returning real numbers from this minute.

What is MCP for blockchain data?

MCP, the Model Context Protocol, is an open standard for connecting AI assistants to outside tools and data. Think of it as a common plug. A tool provider exposes a set of capabilities, the assistant discovers them, and from then on the assistant can call those capabilities in plain conversation.

The Alchemy MCP server is that plug for blockchain data. It exposes Alchemy's API surface as tools your assistant can pick up and use: 168 of them, across 100+ chains. Token balances, NFT ownership, transaction history, portfolio data, prices, gas estimates, transaction simulation, traces, raw RPC calls. If you can do it with an Alchemy API, your assistant can do it through MCP.

Because MCP is a standard and not a plugin for one specific app, you connect once and the assistant handles the rest. There is nothing to install into your project and no SDK to wire up. The server runs at https://mcp.alchemy.com/mcp, your assistant connects to it, and the tools show up.

Every Alchemy API, now something your assistant can call

The useful part is not any single tool. It is that the whole surface is available at once, so your assistant can chain steps together the way you would.

Ask "which wallets supplied more than $1M of USDC to Aave on Base in the last 24 hours," and it can pull the transfers, filter them, and total them up. Ask "what's in this wallet," and it can read balances, token metadata, and NFT holdings and give you one clean summary. Ask "simulate this swap before I send it," and it can run the simulation and tell you what would happen, before any gas is spent.

These are the questions you already ask during a build. Normally each one is a detour: find the endpoint, write the call, parse the response, read it. With every Alchemy API in the assistant, the detour collapses into a sentence.

The server groups tools into three categories. Admin tools manage your Alchemy account and apps. RPC tools cover standard EVM JSON-RPC, Solana RPC, token APIs, transfers, simulation, tracing, debug, and account abstraction. Data tools call Alchemy's REST APIs for NFTs, prices, and multi-chain portfolio data. After you run select_app, the data and RPC tools use that app's credentials for the session. See the full tool list in the MCP server docs.

It works with the assistant you already use

The Alchemy MCP server is not tied to one model or one app. It works with Claude, Cursor, Codex, ChatGPT, and other MCP-compatible clients. You bring the assistant you already build with, and Alchemy connects to it. There is no proprietary model to adopt and nothing to rip out later.

That matters because the right tool is the one already open next to your editor. MCP fits into that, instead of asking you to change how you work.

How do I set it up?

You add the server to your assistant's MCP config once, sign in with your Alchemy account when prompted, and start asking questions. Authentication is OAuth over the hosted server at https://mcp.alchemy.com/mcp. No API key to copy and no local process to run.

For Cursor, add this to ~/.cursor/mcp.json (global) or .cursor/mcp.json (project-level), then restart the editor:

For Claude Code, run:

For Codex:

The first time your assistant calls an Alchemy tool, a browser window opens for OAuth sign-in. After that, tell it to select an app (for example, "select my Alchemy app for Base mainnet") so RPC and Data tools know which credentials to use.

Client-specific steps for Claude Desktop, VS Code Copilot, and other MCP clients are in the Alchemy MCP server documentation. You can also wire the server in one command with alchemy install mcp from the Alchemy CLI.

Once connected, the tools are just there. Open a chat, ask an onchain question in plain language, and your assistant reaches for the right Alchemy API on its own.

When to reach for MCP, the CLI, or Skills

MCP is the fastest path when you want answers in conversation: exploring data, checking state, sanity-testing an idea before you write code. Nothing to install into your repo. Ask and read.

When you want your assistant to take action from the terminal, set up a repo, fund a test wallet, or run a command, the Alchemy CLI is the better fit. And when you want it to write correct Alchemy integration code into an app that ships, Agent Skills load the full API spec so the generated code uses the right endpoints, auth, and pagination the first time.

Most builds end up using all three. MCP is usually where it starts, because asking a question is the lowest-effort way in.

See how the pieces fit together on Alchemy for agents.

Start asking your assistant onchain questions

The same infrastructure that serves 100M+ users across 100+ chains at 99.99% uptime is now something you can talk to. Connect the Alchemy MCP server, ask your assistant what is happening onchain, and watch it answer with real data.