Solana Explorer Launches MCP Server for AI Agent Access to On-Chain Data

The official Solana block explorer shipped a Model Context Protocol (MCP) server on September 17, making its decoded on-chain data directly queryable by AI coding agents. The server is live at explorer.solana.com/mcp, requires no API key, and covers mainnet-beta, devnet, and testnet.

MCP is an open protocol, created by Anthropic, that standardizes how AI models connect to external tools and data sources. Any MCP-compatible agent (Claude Code, Cursor, Windsurf, VS Code, or OpenAI Codex) can now read Solana on-chain state through the Explorer's data pipeline without a custom integration.

What the inspect_entity Tool Returns

The server registers two tools: inspect_entity and ping. Both are read-only; the server cannot sign, send, or alter any on-chain state.

inspect_entity accepts a single base58 input, either a 32-byte account address or a 64-byte transaction signature, and returns the decoded on-chain record for whichever it receives. The tool applies the same IDL decoding and data enrichments the Explorer renders in the browser, so agents receive structured JSON rather than raw account bytes.

For SPL token mints, the tool returns supply, decimals, mint authority, and freeze authority. Programs surface account ownership and metadata. Transactions come back with their instructions and outcomes walked through in sequence. The documentation example shows querying USDC's mint address and receiving supply, decimals, and authority fields in a structured payload.

Two error codes appear in the response's errors[] array: NOT_FOUND when an entity does not exist on the queried cluster, and CURRENTLY_UNSUPPORTED when the account type is recognized but not yet decodable (legacy loader accounts currently fall into this category).

Token market data (live prices, trading volume, and market cap) falls outside what inspect_entity covers. Separate RPC-based and data-provider MCPs handle real-time market queries.

Connecting in One Command

Adding the server to Claude Code takes a single terminal command:

claude mcp add --transport http solana-explorer https://explorer.solana.com/mcp

The MCP page also provides configuration snippets for Cursor, Windsurf, Codex, and VS Code. The setup page recommends adding an instruction block to agent configuration files (AGENTS.md, CLAUDE.md, or Cursor's rules directory) directing agents to call inspect_entity rather than relying on model memory for any Solana address, signature, program, token, or wallet.

The transport is streamable HTTP and stateless, with no session management required.

Two Solana MCP Servers: On-Chain State and Developer Documentation

The Explorer MCP is distinct from a developer documentation MCP that has been available separately at mcp.solana.com. That server provides semantic search across Solana docs, how-to answers, and Rust program security analysis for Anchor and Pinocchio frameworks. The two complement each other: the developer MCP answers questions about how to build on Solana; the Explorer MCP answers questions about what currently exists on-chain.

The pair gives agents a more complete picture of the Solana environment without requiring external RPC credentials for either.

Explorer MCP Extends Solana's AI Agent Toolchain

Solana's on-chain data has previously been accessible to AI agents only through custom RPC integrations or third-party MCP wrappers. The Explorer server moves that access to official infrastructure: the same data layer that powers the browser-based explorer at explorer.solana.com, with no API key required.

Solana's MCP ecosystem has been building since early 2026. Pay.sh added MCP support in June to route AI agent payment approvals through the protocol. The Explorer server extends that pattern to read-only on-chain data, giving agents a reliable source for decoded account and transaction state.