That is the difference from a vendor comparison page: every benchmark's measuring code is in the repository, and pull requests are accepted from anyone including the providers being measured. For an agent, the practical consequence is that `get_benchmark` returns methodology alongside rankings — so a claim can be qualified rather than quoted flat. `query_prom` goes below the published view when the summary is not the cut you need.
The MCP surface of OpenChainBench, a public benchmark project for crypto infrastructure. Each benchmark ships with the harness that produced its numbers, and the server lets an agent list what is measured, read one benchmark's rankings and methodology, and query the underlying metrics store directly.
- `list_benchmarks` enumerates what is measured
- `get_benchmark` returns one benchmark's full rankings, methodology and a paste-ready quote with attribution
- `query_prom` reaches the metrics store behind the published figures
- A resource at `openchainbench://benchmark/{slug}` for reading a benchmark as context
- 94 live benchmark specs across 8 categories — RPCs, Trading, Aggregators, Bridges, Blockchains, RWA, Explorers and NFT APIs — measured by 50 self-contained Go harnesses
- Data is CC-BY-4.0 and every figure carries a citation URL, so a number quoted in a report can be traced
Nothing to install. It is a remote server over Streamable HTTP at `https://openchainbench.com/api/mcp/mcp`, published as io.github.chainbench/openchainbench version 1.0.0, and no key is mentioned. If you want to run the site or a harness yourself, the repository is a pnpm project: `pnpm install`, `pnpm dev` for the site on port 3000, `pnpm validate` to check a benchmark spec against its schema.
One command — npx -y mcp-remote https://openchainbench.com/api/mcp/mcp