dex-data MCP server

Twenty-two tools, no API key and no wallet, in three groups:

• Multi-chain DEX market data — token prices, liquidity depth, pool reserves, best execution venue, liquidity risk, pre-trade slippage and gas costs across BNB Chain, Polygon, Arbitrum, Base, Avalanche and Optimism.

• General-purpose agent utilities — geocoding, reverse geocoding, weather, web search, article/PDF to Markdown, and cryptographic randomness.

• Reference and research data — search autocomplete, public holidays and business days, RSS/Atom feeds, IP geolocation, GLEIF legal-entity lookup and the US Treasury yield curve.

Why this exists

The ledger says it plainly. Of 20,392 paywall challenges recorded, zero were genuine third-party queries — 82.5% were our own monitoring, 17.3% were catalogue crawlers walking every route with no parameters. We were being indexed, not shopped.

Bazaar listings are a bet on demand that does not exist yet. MCP is where agents already are today: Claude Desktop, Claude Code, Cursor, and every other MCP host. This server is the same data, delivered where the users are, and it needs no payment at all for the free tier — which removes the one barrier that a browsing agent cannot cross on its own.

Related MCP server: AgentData MCP Server

Install

Claude Code

claude mcp add dex-data -- npx -y dex-data-mcp

Claude Desktop / Cursor — add to claude_desktop_config.json:

{
  "mcpServers": {
    "dex-data": {
      "command": "npx",
      "args": ["-y", "dex-data-mcp"]
    }
  }
}

No API key, no wallet, no signup. The free tier answers every tool for a daily allowance; after that the endpoints fall back to x402 micropayments.

Tools

DEX market data

General-purpose agent utilities

Not a change of theme — these were chosen the same way everything else here was. Reading USDC receipts across 1,062 x402 seller wallets on Base ranked what actually gets paid for, and DEX data was not near the top of that list. Forward geocoding was (56 paying wallets), then weather, then article-to-Markdown, then randomness. These are the answers to that data.

Reference and research data

Chosen the same way, but from a corrected reading. The category totals in the original scan were double-counted — the biggest x402 sellers carry 10-16 of the 16 category tags each, so every category reported nearly the whole market's revenue. Seller-level rows are clean, and every tool below is something a wallet took real USDC for during the sampled window.

Local

Pairings worth knowing: geocode then get_weather turns a place name into a forecast, search then url_to_markdown turns a question into readable source text, and read_feed then url_to_markdown turns a feed into full articles.

Free tier

Every tool answers free within a daily per-caller allowance — no wallet, no signup. The remaining quota is returned on X-FreeTier-Remaining. Beyond it, calls fall back to x402 micropayments (USDC on Base).

Every response carries the liquidity backing the number and a confidence rating. A quote with no depth behind it is refused rather than returned — a dust-pool price is worse than no price when an agent may trade on it.

Honest limits

• Depth and prices cover the venues this API indexes; a deeper pool may exist on a DEX not covered here.

• v3 figures distinguish custody TVL from tradeable depth at the current price. They are not the same number and are not labelled as if they were.

• get_slippage is pool-level price impact: it excludes gas, MEV and multi-hop routing, and flags v3 estimates that cross the active tick band.