Love what you've built here — evm-mcp-server is the right foundation for AI agents working with on-chain systems.
One gap I keep running into: the server uses hardcoded public endpoints (LlamaRPC, dRPC, PublicNode, chain-official RPCs) but there's no way for the agent to know which of those endpoints are actually healthy or fast at a given moment. A slow or stale RPC silently degrades the agent's results.
OpenChainBench continuously benchmarks those exact providers across 40+ chains and exposes the data through a keyless MCP server:
- MCP endpoint:
https://openchainbench.com/api/mcp/mcp (Streamable HTTP, no auth required)
- Tools:
list_benchmarks, get_benchmark (current rankings, p50/p99 latency, success rate), query_prom (raw Prometheus)
- Repo: https://github.com/ChainBench/OpenChainBench
The agent workflow with OCB:
- Agent calls OCB MCP to get current RPC latency rankings for the target chain
- Agent picks the fastest/most reliable endpoint from the live data
- Agent executes the EVM call via evm-mcp-server using that endpoint
Not asking for a code change — flagging it as a companion tool for users who want RPC visibility, the same way the security scanning tool was flagged in #37. Happy to add evm-mcp-server as a usage example in the OCB docs if useful.
Love what you've built here — evm-mcp-server is the right foundation for AI agents working with on-chain systems.
One gap I keep running into: the server uses hardcoded public endpoints (LlamaRPC, dRPC, PublicNode, chain-official RPCs) but there's no way for the agent to know which of those endpoints are actually healthy or fast at a given moment. A slow or stale RPC silently degrades the agent's results.
OpenChainBench continuously benchmarks those exact providers across 40+ chains and exposes the data through a keyless MCP server:
https://openchainbench.com/api/mcp/mcp(Streamable HTTP, no auth required)list_benchmarks,get_benchmark(current rankings, p50/p99 latency, success rate),query_prom(raw Prometheus)The agent workflow with OCB:
Not asking for a code change — flagging it as a companion tool for users who want RPC visibility, the same way the security scanning tool was flagged in #37. Happy to add evm-mcp-server as a usage example in the OCB docs if useful.