Integration

Use OpenChainBench from your AI assistant

OpenChainBench ships an MCP server. Point Claude Desktop, Cursor, ChatGPT, or any MCP-capable client at one URL and live crypto-infra benchmarks become a first-class tool for your model.

Server URL

https://openchainbench.com/api/mcp/mcp

Transport: streamable HTTP · Auth: none · Rate limit: 60 req/min/IP

Install in Claude Desktop

Add this block to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or the equivalent on your OS, then restart the app.

{
  "mcpServers": {
    "openchainbench": {
      "url": "https://openchainbench.com/api/mcp/mcp"
    }
  }
}

Once connected, the three tools appear under the 🔌 icon in the chat input. Ask Claude “which crypto aggregator is fastest on Base today?” and it will call get_benchmark and cite the live number.

Install in Cursor

Settings → MCP → Add server → paste the URL above. Or drop the same JSON block into ~/.cursor/mcp.json:

{
  "mcpServers": {
    "openchainbench": {
      "url": "https://openchainbench.com/api/mcp/mcp"
    }
  }
}

III

Any other MCP client

Continue, Zed, Cline, Goose, custom agents on the MCP SDKs: all accept the same URL with the streamable-HTTP transport. SSE is intentionally disabled. Anything else, raw curl works:

curl -s -X POST https://openchainbench.com/api/mcp/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"get_benchmark","arguments":{"slug":"aggregator-head-lag","chain":"base"}}}'

Surface

What's exposed

Tool

list_benchmarks()

Flat index of every live benchmark with current value, leader, category, and citation URL. The discovery call your agent runs first.

Tool

get_benchmark(slug, chain?, region?)

Full detail for one bench: rankings, sparkline, headline, paste-ready citation quote, methodology. Filter by chain or region when the spec declares them.

Tool

query_prom(query, windowSec?, steps?)

Direct PromQL passthrough, scoped to published benchmark metric namespaces. Allowlist enforced: wallet inventories and operational metrics are refused.

Resource

openchainbench://benchmark/{slug}

18 benchmarks each exposed as an MCP resource (Markdown + JSON in the same read). Pin one into your agent's context as a long-lived document instead of re-fetching every turn.

Examples

Ask your agent

· Which crypto data aggregator is fastest on Base today?
· How much does it cost to bridge $300 USDC from Solana to Arbitrum?
· Plot bridge fees over the last 24 hours.
· Compare Mobula and GeckoTerminal on network coverage.
· Cite the latest finality lag for BNB Chain in a paragraph I can paste.

Safety

Open by design

The server is read-only. There are no admin actions, no mutating tools, no upstream secrets reachable through the wire. PromQL queries are scoped to the metric namespaces we publish, so the public endpoint can't be used to walk the underlying Prometheus. Rate limits, body caps, batch rejection and SSE disable are enforced at the route. Source is on GitHub.

Trouble? Drop a note on /contribute or open an issue. We watch the channel.