Arb DEX

Live cross-DEX crypto prices for your AI agent— per-venue pool price, pool liquidity and the gross cross-venue spread on 6 EVM chains, read straight from on-chain pool state.

Chains:BSC · Polygon · Arbitrum · Base · Avalanche · Optimism. Venues: PancakeSwap (v2 + v3), Uniswap v3, SushiSwap, QuickSwap, Biswap, ApeSwap, BaseSwap, Trader Joe, Pangolin — every v2-style pooland every v3 fee tier separately, because a $12k 1% pool and a $19M 0.01% pool are not the same quote.

Nothing is modelled, estimated or backfilled. Every payload states its own block number and carries its own scope note, so an agent that quotes a figure also has the caveats attached to it.

Works with no API keyagainst a free hourly public snapshot.Docs·npm

{ "mcpServers": { "arb-dex": { "command": "npx", "args": ["-y", "arb-dex-mcp"], "env": { "RAPIDAPI_KEY": "your-rapidapi-key-here" } } } }

Restart Claude Desktop; the six tools appear under the connectors icon.Drop theenvblock entirely to run keyless— the server still starts and the free-snapshot tools still answer.

claude mcp add arb-dex --env RAPIDAPI_KEY=your-rapidapi-key-here -- npx -y arb-dex-mcp

claude mcp add arb-dex -- npx -y arb-dex-mcp

~/.cursor/mcp.json(global) or.cursor/mcp.json(per project):

{ "mcpServers": { "arb-dex": { "command": "npx", "args": ["-y", "arb-dex-mcp"], "env": { "RAPIDAPI_KEY": "your-rapidapi-key-here" } } } }

Cursor → Settings → MCP shows the server and its tools once the file is saved.

Same three facts: commandnpx, args["-y", "arb-dex-mcp"], transportstdio, optional envRAPIDAPI_KEY.

Real prompts, and the real shape that comes back. Payloads below were measured live on2026-08-15; they are trimmed for width but nothing is invented.

1. "What is WBNB/USDT trading at on every BSC venue right now?"

get_pricesreads every pool holding the pair — v2 pairs and each v3 fee tier separately — at one stated block:

{ "pair": "WBNB/USDT", "network": "bsc", "chainId": 56, "blockNumber": 116151268, "pricesByVenue": [ { "venue": "pancake", "surface": "v2", "feeBps": 25, "price": 611.7347, "tvlUsd": 56834814.36 }, { "venue": "biswap", "surface": "v2", "feeBps": 10, "price": 610.8278, "tvlUsd": 415733.94 }, { "venue": "apeswap", "surface": "v2", "feeBps": 20, "price": 611.1244, "tvlUsd": 3659.06 }, { "venue": "pancakeV3:1", "surface": "v3", "feeBps": 1, "price": 610.5684, "tvlUsd": 18596810.32 }, { "venue": "pancakeV3:5", "surface": "v3", "feeBps": 5, "price": 610.6531, "tvlUsd": 4941099.44 }, { "venue": "pancakeV3:25", "surface": "v3", "feeBps": 25, "price": 610.6304, "tvlUsd": 52534.35 }, { "venue": "pancakeV3:100", "surface": "v3", "feeBps": 100, "price": 609.1025, "tvlUsd": 12206.21 } ], "bestBuy": { "venue": "pancakeV3:100", "price": 609.1025 }, "bestSell": { "venue": "pancake", "price": 611.7347 }, "midSpreadBps": 43.22, "crossDex": { "grossSpreadBps": 0, "grossUsd": 0, "optimalInput": { "amount": 0, "token": "WBNB", "usd": 0 }, "buyVenue": "-", "sellVenue": "-" }, "liquidity": { "venues": 7, "totalTvlUsd": 80856857.67 }, "source": "rpc" }

Read the two spread numbers against each other.The raw mid spread is43 bps— and the gross capturable spread is0. The 609.10 quote lives in a $12k pool; the size that would actually clear it moves the price past the gap before you get there. A tool that reported only the 43 bps would be handing an agent a number it cannot trade. This one reports both, andoptimalInputis where the honesty lands.

2. "Show me the cross-DEX spreads on Base — are any actually capturable?"

get_spreadssweeps a whole chain and ranks bygross USD at the optimal size, not by headline basis points:

{ "network": "base", "chainId": 8453, "scannedPairs": 11, "opportunities": [], "found": 0, "filters": { "minSpreadBps": 10, "minVenueTvlUsd": 1000, "minGrossUsd": 0.01, "limit": 5 }, "ranking": "gross USD at the optimal trade size, NOT raw spread — a large spread with a tiny optimal size is not an opportunity", "scope": "GROSS cross-venue spread from live pool state, BEFORE gas, MEV and any slippage beyond the optimal size. Not a profit estimate and not trade advice. Venues below the liquidity floor are excluded because a spread against a dust pool is an artefact, not an opportunity.", "elapsedMs": 2847 }

found: 0is a real answer and it is the common one.Eleven pairs scanned, nothing cleared the floor. Venues under $1,000 TVL are dropped outright. When rowsdocome back, each carriescapturable,warningandshallowestSideTvlUsdso a big basis-point number cannot mislead on its own. This tool will tell your agent there is nothing there — which is the whole point of asking it.

3. "How much history does arb-dex actually have, and for which chains?"

get_history_summarysizes the archivebeforeyou query it:

{ "rows": 133, "rowsWithPairDetail": 103, "rowsByEra": { "digest-totals-only": 30, "top-list-pairs": 5, "full-sweep": 98 }, "pairsTracked": 107, "firstAt": "2026-08-10T17:35:09.355Z", "lastAt": "2026-08-15T20:54:34.417Z", "spanHours": 123.32, "chainsSeen": ["arbitrum", "avalanche", "base", "bsc", "optimism", "polygon"], "pairs": [ { "chain": "polygon", "pair": "WBTC/USDC", "observations": 103, "qualifiedObservations": 39 }, { "chain": "bsc", "pair": "BTCB/USDT", "observations": 99, "qualifiedObservations": 15 }, { "chain": "arbitrum", "pair": "ARB/USDC", "observations": 98, "qualifiedObservations": 0 } ] }

Coverage is only what was measured. A gap stays a gap —rowsByErasays how much detail each era of rows carries, andARB/USDChaving 98 observations but0 qualifiedis the archive telling you that pair has never once cleared the spread floor.

The two history tools read the service's own measurement archive, so they answer the question the live tools cannot: whether a dislocationpersistedor was a single sample. Sampling is roughly hourly, and gaps are never interpolated or backfilled. Callget_history_summaryfirst to see what span exists before asking for a window.

- Spreads are gross— before gas, MEV and slippage beyond the optimal size. Not a profit estimate and not trade advice.
- It never fabricates a row.Without a key,get_chains,get_pairsandget_spreadsanswer from the free public surface and each carries alimitationfield naming exactly what a key would add.get_prices,get_history_summaryandget_historyreturn an explicit key-required error with the signup link rather than a thinner answer dressed up as a full one.
- It does not execute trades, hold funds, or touch a wallet. It is read-only market data.
- This package ships no credentials of any kind.The key is yours and stays in your config.

The paid tools call the API through RapidAPI usingyour ownkey.

Five of the six tools work on thefreetier. Only the per-pair history series is gated:

Callingget_historybelow PRO returns an explicittier_requirederror naming the plan and the upgrade URL — it does not fail silently or return an empty series.

Set these in your MCP client'senvblock (see the configs above)..env.exampleships in the package and documents the same variables for local runs from a clone.

The test suite is not in the npm tarball — run it from a clone:

git clone https://github.com/donnywin85/arb-dex-mcp.git cd arb-dex-mcp && npm install npm run selftest # keyless: exercises the free fallbacks RAPIDAPI_KEY=... npm run selftest # keyed: exercises the paid routes

The test spawns the server over stdio and calls every tool against thereal production API— nothing is mocked. It asserts on live values (block number, per-venue prices, scanned-pair counts), so a run that passes is evidence the data path works end to end.

- npm:https://www.npmjs.com/package/arb-dex-mcp
- Docs:https://donnywin85.github.io/arb-dex-mcp/
- Official MCP Registry:io.github.donnywin85/arb-dex-mcp
- TensorBlock MCP Index:https://www.tensorblock.co/mcp/servers/github-donnywin85-arb-dex-mcp-7dd9a70c
- Source:https://github.com/donnywin85/arb-dex-mcp
- The API behind it:https://rapidapi.com/donnydev/api/multi-chain-dex-prices-liquidity

Access blockchain transaction data in natural language, providing human-readable descriptions for AI assistants.