MCP to Talk to Blockchain Data

Bitquery has launched an MCP server that lets AI agents in Claude, ChatGPT, Cursor, and VS Code query on-chain blockchain data for crypto investigations and market intelligence. The tool indexes every DEX trade on eight chains down to the second, including tokens not listed on CoinGecko, and offers one-shot workflows for money-flow tracing, risk screening, and address dossiers. Bitquery claims it performs the same job as costly investigation tools for a fraction of the price, enabling plain-English queries that return evidence-backed answers.

On-chain market intelligence and crypto investigations for AI agents — inside Claude, ChatGPT, Cursor and VS Code. Buttons target this host automatically — production mcp.bitquery.io or staging. Claude Code has no deep-link yet, so its button copies the install command to your clipboard; for ChatGPT and Codex follow the manual steps below. 📖 Full installation guide: docs.bitquery.io/docs/usecases/MCP https://docs.bitquery.io/docs/usecases/MCP/ Every DEX trade on 8 chains, indexed down to the second — including tokens that never reach CoinGecko. From one address to the whole story — money flows, labels, risk verdicts, flow maps. Most crypto investigation tools cost thousands of dollars a month. Bitquery MCP does the same job inside Claude or ChatGPT: paste a wallet address, ask your question in plain English, and get the answer back — with the evidence. In the video above we point it at a single Tron wallet and let it work: it pulls the full transaction history, labels the counterparties, tells known exchanges apart from unknown wallets, traces where the money actually moved, and writes a clear verdict — what the wallet is doing, how risky it looks, and the leads worth chasing next. The money-flow map at the end is generated in the same run. money flow prompt risk screen prompt address report prompt entity report prompt address labels — entity, category, CEX deposit / hot wallet, mixer, gambling, scam, darknet and token-clone labels btc related addresses — common-input-ownership clusteringThese are one-shot workflows MCP prompts/ that chain several tools and return a chart, table or diagram. Pick them from your client's prompt / command menu: money flow Trace funds from an address or transaction across hops, label counterparties, stop at exchanges / mixers → money-flow diagram + edge table. risk screen AML / compliance screen of an address or transaction mixer / sanctioned / scam / darknet → LOW / MEDIUM / HIGH verdict + evidence + exposure diagram. address report Full dossier for one address: activity stats, recent transfers, labels, and Bitcoin likely same-owner addresses. entity report Resolve a labeled entity → its addresses → activity and connections. The prompts are built from per-chain tracing tools your agent can also call directly — the same family exists for each traced chain eth , base , arbitrum , matic , optimism , tron , solana , btc : address flow summary — one-call triage of any hop: transfer counts, top senders & receivers, and a computed role — flow edges — money-flow graph edges source → target, amount, token , ready to render as a diagram. trace next hop — "where did the bulk of the funds go": outgoing flow aggregated by counterparty, largest first. trace dominant path — auto-follows the largest edge hop by hop, up to 5 hops in a single call. btc tx flow — full inputs and outputs of a Bitcoin transaction with change outputs flagged, so you follow the real payment. address labels / addresses by label — the label directory both directions: identify one address, or pull every address tagged Investigate with FIFO, LIFO, or any methodology you prefer — the agent applies it across the raw transfer data. Transfer tracing covers: If you work in compliance and AML, investigate fraud, or just follow the money on-chain, this gets you from an address to the whole story in about a minute. Bitquery indexes every DEX trade on Ethereum, Arbitrum, Base, Polygon, Optimism, BNB Chain, Tron and Solana down to the second — so the answers cover freshly-launched memecoins and long-tail tokens that never appear on CoinGecko or CoinMarketCap. You ask in plain English; the agent picks the right tool: trending tokens pair ohlcv profitable traders by token and trader positions trader profile and trader activity find tokens accumulating traders by token ; profitable traders by token ranks by realized + unrealized PnL instead tx trades eth token holders also Polygon, Arbitrum, Base, Optimism execute sql paid plans token chart Resolve a token and render a candlestick / line price chart with a data table. token dashboard 360° view of one token: price chart + supply / FDV + top & smart-money traders + scam check. trader dashboard One wallet: profile stats + holdings / PnL positions + activity time-series. compare assets Overlay 2–5 assets on one normalized %-change price chart + stats table. token scout Trending / gainers + momentum + accumulation → ranked pivot of promising tokens. market movers Chain leaderboard: top gainers + most-traded → sortable table + bar chart. arbitrage scan One asset across chains → price-spread % table + bar chart. Every prompt and example above is built from individual MCP tools your agent can also call directly. The catalog groups into six families: Discovery find tokens resolves a name or ticker with wildcards to contract + chain; find currencies , token chains and token dex venues map an asset across chains and DEXs; chain capabilities returns the per-chain tool matrix. Prices & charts token price / token ohlcv , pair ohlcv USD- or quote-denominated , cross-chain currency aggregates for majors, plus supply, market cap and pool recent trades . Candles from 1 second to 1 hour. Traders trending tokens volume / gainers / losers / most-volatile , top traders by token, pair or network, profitable and accumulating traders , and per-wallet trader profile / positions / activity . Tracing A per-chain family for 8 chains: address flow summary , flow edges , trace next hop , trace dominant path , transfers in/out , tx transfers , token holders , plus contract find events / find calls . Labels address labels for one address, labels for addresses for a batch, addresses by label for every address tagged X, find label values to discover what exists — entities, CEX deposit & hot wallets, mixers, gambling, scams, darknet, token clones. Raw SQL execute sql runs read-only ClickHouse over 30 days of per-second DEX trades plus the label directory; transfers raw sql does the same on each chain's transfer data paid plans . The full tool list, parameter shapes, and supported networks are returned by the server itself when your client calls tools/list — so the AI always sees the up-to-date catalog without you reading docs. Per-chain coverage varies slightly e.g. Solana exposes solana find instructions , Bitcoin adds btc related addresses clustering ; chain capabilities is the authoritative matrix. The underlying dataset is the same Bitquery you already know. The difference is the interface: Install Bitquery as a custom connector in your AI client. No config file editing required. Watch the walkthrough: https://mcp.bitquery.io Add Bitquery to Claude Code with one command the Add to Claude Code button above copies it to your clipboard : claude mcp add --transport http bitquery https://mcp.bitquery.io Append --scope user to enable it across every project. On first use, run /mcp inside Claude Code or claude mcp login bitquery and complete the browser OAuth login with your Bitquery account. https://mcp.bitquery.io https://mcp.bitquery.io Add Bitquery to your Codex MCP servers with the CLI: codex mcp add bitquery --url https://mcp.bitquery.io Or edit ~/.codex/config.toml directly: mcp servers.bitquery url = "https://mcp.bitquery.io" On first call, Codex opens a browser for OAuth login with your Bitquery account. Add a new MCP server with URL: https://mcp.bitquery.io It is a hosted Model Context Protocol MCP server at https://mcp.bitquery.io that exposes Bitquery's on-chain blockchain data as MCP tools. AI agents like Claude, ChatGPT, Cursor, and VS Code can call these tools directly during a conversation, without you having to write GraphQL queries by hand. Yes. The server ships guided investigation prompts: money flow traces funds across hops and stops at exchanges or mixers, risk screen returns a LOW / MEDIUM / HIGH AML verdict with evidence, address report builds a full dossier for one address, and entity report resolves a labeled entity to its addresses and activity. Transfer tracing covers Ethereum, Polygon, Arbitrum, Base, Optimism, Tron, Bitcoin and Solana, backed by millions of address labels. Any client that supports remote MCP servers over Streamable HTTP. This includes Claude Desktop and Claude.ai web, with Plus/Pro/Max , ChatGPT Plus/Pro/Business with custom connectors enabled , Cursor, VS Code with MCP enabled, and custom agents built with the Anthropic or OpenAI SDKs. DEX trading tools cover Ethereum, Arbitrum, Base, Polygon, Optimism, BNB Chain, Tron and Solana, with candles down to 1-second granularity; transfer-tracing tools cover Ethereum, Polygon, Arbitrum, Base, Optimism, Tron, Bitcoin and Solana. Tools span token prices, OHLCV candles, trending tokens, top / profitable / accumulating traders, trader profiles and positions, DEX pairs, transfers, the address-label directory, and raw SQL access where the user has a paid plan. Authentication uses OAuth 2.1 : the AI client opens a browser to your Bitquery account, you approve access, and tokens refresh automatically for ~30 days. Access is tied to your Bitquery account, so revoking it there immediately disables the connection. The MCP server is built for AI agents. Instead of writing a GraphQL query, the agent picks an MCP tool for example token ohlcv or top traders by token and the server runs the underlying query on Bitquery for you. The data source is the same; the interface is conversational. Quotas, plans, and billing are shared with your regular Bitquery account. There is a free tier inherited from your Bitquery account. Heavier workloads — high request rates, large historical scans, or execute sql — require a paid Bitquery plan. The MCP server itself does not add a separate charge.