If you're profiling Linux workloads on Arm64, this server plugs three optimization steps directly into Claude. It parses perf report stdio output into ranked hot symbols, searches a curated knowledge base of 110 NEON intrinsics using semantic similarity to suggest the right vmlaq_f32 or vaddq_u8 for your hot loop, and audits Python requirements files for packages that lack arm64 wheels or need special handling. All three tools run offline in a Docker container over stdio. The NEON retrieval hits the right intrinsic in the top 3 results 93% of the time on the eval set. Useful when you're porting x86 SIMD code or chasing down dependency blockers before deploying to Graviton.
claude mcp add --transport stdio jean-johnson-zwix-arm-code-mcp uvx arm-code-mcp