Model Context Protocol server

Point any MCP-capable agent at mcp.avee.tech and it reads the same live DEX index this terminal draws: pairs, tokens, prices, liquidity and the raw swap tape across every indexed chain. No account, no key, no SDK.

Streamable HTTPSSE (legacy)Anonymous 10 req/sWith API key 30 req/s

Endpoint

https://mcp.avee.tech/mcp

Legacy SSE endpoint

https://mcp.avee.tech/sse

An API key is optional and only raises the rate limit. Send it as X-API-Key: YOUR_KEY or Authorization: Bearer YOUR_KEY. Ask for one at connect@avee.tech.

Connect your client

Claude Code

One command. Drop --header if you have no key.

terminal

claude mcp add --transport http avee https://mcp.avee.tech/mcp \
  --header "X-API-Key: YOUR_KEY"

Claude Code / Codex — project file

Commit this next to the repo so every agent on the team gets the same tools.

.mcp.json

{
  "mcpServers": {
    "avee": {
      "type": "http",
      "url": "https://mcp.avee.tech/mcp",
      "headers": { "X-API-Key": "YOUR_KEY" }
    }
  }
}

Claude Desktop / claude.ai

Settings → Connectors → Add custom connector, then paste the endpoint.

connector URL

https://mcp.avee.tech/mcp

Cursor

Global at ~/.cursor/mcp.json, per project at .cursor/mcp.json.

~/.cursor/mcp.json

{
  "mcpServers": {
    "avee": {
      "url": "https://mcp.avee.tech/mcp",
      "headers": { "X-API-Key": "YOUR_KEY" }
    }
  }
}

VS Code — GitHub Copilot

VS Code keys the block "servers", not "mcpServers".

.vscode/mcp.json

{
  "servers": {
    "avee": {
      "type": "http",
      "url": "https://mcp.avee.tech/mcp"
    }
  }
}

ChatGPT

Settings → Connectors → Create, MCP server URL, no authentication.

connector URL

https://mcp.avee.tech/mcp

A client that only speaks stdio

Bridge the remote endpoint through mcp-remote.

mcpServers entry

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

No client at all

The transport is plain JSON-RPC over HTTP — curl works.

terminal

curl -sN https://mcp.avee.tech/mcp \
  -H 'Content-Type: application/json' \
  -H 'Accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Tools

Tool	What it returns	Required arguments
get_pair_list	Liquidity pairs with filters for chain, DEX, protocol, liquidity band and free text.	— (all optional)
get_pair	One pair in full, optionally with 5m/1h/6h/24h aggregated statistics.	address, chain_id
get_pair_base_info	The cheap slice of a pair: liquidity USD, fee and ticker.	address, chain_id
get_base_token	Token symbol, decimals, supply and price.	address, chain_id
get_best_price_usd	The best USD price for a token across every pair that holds it.	address, chain_id
get_order_book	The raw swap tape for a pair — buys, sells, mints and burns, cursor paginated.	address, chain_id
get_order_book_count	How many transactions match a tape filter, without pulling the tape.	address, chain_id

Questions

Do I need an API key to use the avee MCP server?

No. mcp.avee.tech answers anonymously at 10 requests per second. A key raises that to 30 requests per second and nothing else — the data is identical.

Which transport does the avee MCP server speak?

Streamable HTTP at /mcp, which is what every current MCP client wants. The legacy SSE transport stays available at /sse for older clients.

Which chains does it cover?

Every chain the terminal indexes — Ethereum, BSC, Base, Arbitrum, Polygon, Avalanche, Optimism, Fantom, TON, Solana, Hyperliquid and HyperEVM. Tools take a numeric chain_id.

Is the data live?

Yes. The MCP server reads the same index as the terminal, so prices, liquidity and the swap tape are as fresh as the block that produced them.

What happens when I exceed the rate limit?

The tool call fails with a rate-limited error rather than a partial result, so an agent can back off and retry without reasoning over half a page.