Skip to main content

Bitquery MCP Server — Blockchain Data for AI Agents

Point your AI client at https://mcp.bitquery.io and ask Bitquery's blockchain dataset questions in plain English. Works with Claude, Cursor, ChatGPT, Claude Code, and anything else that speaks MCP.

You get outlier-filtered DEX trades, transfers, OHLC candles, market cap, and wallet history across Solana, Ethereum, BSC, Base, Arbitrum, Optimism, Polygon, Tron, and Robinhood Chain. Same data behind the tracing side: origin of funds, cross-chain movement, wallet clusters.

Endpointhttps://mcp.bitquery.io
AuthOAuth 2.1, browser sign-in, no manual tokens
PricingFree tier with a Bitquery account. Beyond that it meters against your existing plan.
SupportTelegram · support.bitquery.io · support@bitquery.io

What It Is​

A hosted Model Context Protocol endpoint sitting in front of Bitquery's production trading and transfers dataset. It's the same data that powers the GraphQL APIs, the IDE, the Kafka streams, and the TradingView feeds.

You ask a question. The agent does the lookup. You get structured rows back. You never write a query, sync a schema, or think about rate limits.

What your agent can actually reach:

  • Billions of swap-level trade rows across Solana, Ethereum, BSC, Base, Arbitrum, Optimism, Polygon, and Tron.
  • Transfer-level data for tracing origin of funds, following money across chains, and pulling apart wallet clusters.
  • Tracing and forensics workflows: AML/KYC scoring, payment paths, phishing investigations, stablecoin and bridge monitoring, sanctions screening, forensic reports.
  • OHLC candles already built at 1-minute, 5-minute, hourly, and daily intervals.
  • Market cap, fully diluted valuation, and circulating supply attached to every row.
  • Bitquery's outlier filter, applied automatically, so wash-traded pools and noisy pairs drop down the ranking.
  • No ETL, no schema sync, no API client to write.

How It Works​

┌──────────────────┐    MCP (JSON-RPC)    ┌─────────────────────┐   queries   ┌──────────────────────┐
│  Claude / Cursor │  ─────────────────►  │  mcp.bitquery.io    │  ────────►  │  Bitquery production │
│  ChatGPT / Code  │  ◄── tool results ── │  (OAuth 2.1)        │  ◄── rows ──│  trading dataset     │
└──────────────────┘                      └─────────────────────┘             └──────────────────────┘
  1. Your client connects to https://mcp.bitquery.io over the MCP transport.
  2. OAuth 2.1 handles sign-in. A browser window opens once, then the client caches a refresh token for about 30 days and renews it quietly in the background.
  3. The server exposes a small set of tools. The model uses them to figure out what's in the dataset and pull rows.
  4. Access is read-only. The agent can't write, delete, or modify anything, even if you ask it to.
  5. When you ask for "clean" volume, Bitquery's price-index ranking is applied for you.

Because it's standard MCP, any compatible client works without special handling: Claude Desktop, Claude Code, Cursor, ChatGPT, VS Code, or something you built yourself.

What Data You Get​

Two datasets are live. The agent picks between them based on what you ask.

Trading covers all nine supported chains, with two views of the same underlying trades:

  1. Per-trade rows: every swap, with price, USD amounts, supply snapshot, trader wallet, and transaction hash.
  2. Pre-built candles: OHLC, volume, and price averages bucketed at 1-second, 1-minute, 5-minute, hourly, and daily intervals.

Tracing handles fund-flow and compliance questions:

  1. AML/KYC risk scoring, plus sanctions and OFAC screening.
  2. Payment tracing from source to destination, including hops, DEX swaps, and bridges.
  3. Investigations: phishing and fraud, money-laundering patterns, CEX deposit clustering.
  4. Wallet clustering, stablecoin movement monitoring, and forensic or SAR-style reporting.

More on trading:

More on tracing:

Under the Hood​

The server is built on the open-source mcp-clickhouse project. Your agent's natural-language requests get turned into read-only queries against Bitquery's production cluster, and you never have to see, write, or debug one.

Writes, deletes, and mutations are rejected at the server, so there's nothing you can break by poking around.

Install​

Claude (Desktop & Web)​

  1. Open Settings → Connectors → Add custom connector.
  2. Fill in:
    • Name: Bitquery Trading Data
    • URL: https://mcp.bitquery.io
  3. Click Add.

Add custom connector: Name and Remote MCP server URL fields

You don't need to pick tools by hand in the chat UI. Ask in plain language, for example "get me the top Solana tokens by 24h volume", and Claude calls Bitquery when the request fits.

ChatGPT​

  1. Settings → Connectors (requires Plus, Pro, or Business).
  2. Add connector → Custom connector.
  3. URL: https://mcp.bitquery.io.
  4. Authenticate with your Bitquery account when prompted.

Cursor​

Easiest path:

  1. Settings → MCP → Add new MCP server.
  2. URL: https://mcp.bitquery.io.

Or add it to .cursor/mcp.json directly:

{
  "mcpServers": {
    "bitquery": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.bitquery.io/mcp"]
    }
  }
}

Restart Cursor, or reload the MCP configuration, so the server gets picked up.

Claude Code (terminal)​

claude mcp add bitquery -- npx -y mcp-remote https://mcp.bitquery.io/mcp

VS Code and other MCP clients​

Add a new MCP server with the URL https://mcp.bitquery.io. The client handles OAuth on its own.

First Connection and Permissions​

Once it's configured, Bitquery Trading Data shows up in your client's connector list:

Bitquery MCP listed among custom connectors

The first time a tool runs, the client asks you to connect and approve:

Tool run requesting Connect for Bitquery

You're prompted to sign in to Bitquery. A free account works:

Bitquery sign-in to authorize the application

Then authorize the client (Claude, Cursor, whatever you're using) to act on your account:

Authorize Bitquery access for the requesting application

In Cursor and similar clients, you choose whether tool calls always need approval or run automatically:

Tool permission options such as Always allow

OAuth tokens are cached for roughly 30 days and refreshed for you, so re-authenticating is rare.

What You Can Ask the Agent​

Most trader and investigator questions fall into one of these shapes. Type it into the chat and let the agent work out the rest.

  • Token discovery and trending: "Top 10 Solana tokens by USD volume in the last 24h, skip wash-traded pools."
  • Trader PnL and wallet analytics: "Pull every trade for wallet 7xKX… in the last 7 days. Compute realised PnL per token."
  • OHLC charts: "1-minute OHLC for the WIF/USDC pool on Raydium for the last 6 hours."
  • Market cap and FDV monitoring: "Which Base tokens crossed $10M market cap in the last 24h?"
  • Wash-trade filtering: saying "only clean volume" is enough to trigger Bitquery's outlier ranking.
  • Launchpad pulse: "How many new tokens launched on Pump.fun in the last hour vs the 24h average?"
  • Cross-chain market overview: "For each chain, show 24h DEX volume, trades, and unique traders."
  • Sniping and copy-trading research: prototype a signal here before you commit to a Kafka or gRPC stream.
  • Slippage and liquidity: derive realised slippage and effective depth from per-trade rows.
  • AML/KYC scoring: "Score wallet 0x… for AML risk. Check wallet age, mixing signals, and compliance flags."
  • Payment tracing: "Trace $500K USDT from this hot wallet to the final destination. Show every hop."
  • Phishing and fraud: "2 ETH were stolen from this wallet. Trace where the funds went across swaps and bridges."
  • Wallet clustering: "Cluster wallets that sent to or received from 0x… by likely control."
  • Sanctions and reporting: "Screen this wallet against OFAC and known-risk entities, then draft a forensic summary."

For end-to-end trading workflows, see the worked examples. For tracing patterns and prompting habits, see Tracing on the MCP. For more trading patterns, see What you can do with it.

Frequently Asked Questions

Is the Bitquery MCP server free?

A free Bitquery account is enough to start. Heavy usage meters against the same plan as the GraphQL API.

Do I need to write SQL to use the MCP?

No. You ask in plain English and the agent works out the rest. The server is read-only, so the agent can't delete, insert, or modify anything.

Which chains and DEXs does the Bitquery MCP cover?

Solana (Raydium, Pump.fun, Meteora, Orca and others), Ethereum (Uniswap, PancakeSwap, Curve), BSC (PancakeSwap, FourMeme), Base, Arbitrum, Optimism, Polygon (Polymarket), and Tron. The Trading overview page has the full DEX list per chain.

Can I use Bitquery MCP for AML, forensics, or fund tracing?

Yes. Tracing covers AML/KYC scoring, payment tracing, phishing investigations, wallet clustering, stablecoin and bridge monitoring, sanctions screening, and forensic reporting. Ask in plain English. The Tracing overview has example prompts.

How fresh is the data on the Bitquery MCP?

Near real time. The dataset updates as blocks are processed across every supported chain.

How do I install the Bitquery MCP in Claude, Cursor, or ChatGPT?

Open Settings → Connectors (or MCP Servers), add https://mcp.bitquery.io as a custom connector, and sign in with your Bitquery account when prompted.

Can I self-host the MCP server?

The underlying server is open-source mcp-clickhouse. Bitquery runs the production deployment with OAuth and our data. You can run your own against your own database.

How far back does MCP data go?

Multiple years for major chains. Tell the agent the window you care about, for example 'in the last 7 days', and it bounds the lookup efficiently.

References​