diff --git a/python/pyproject_cpu.toml b/python/pyproject_cpu.toml
index 38ae6490d..e28e6d99b 100644
--- a/python/pyproject_cpu.toml
+++ b/python/pyproject_cpu.toml
@@ -97,6 +97,9 @@ dev = ["sglang[test]"]
 "Homepage" = "https://github.com/sgl-project/sglang"
 "Bug Tracker" = "https://github.com/sgl-project/sglang/issues"
 
+[project.scripts]
+sglang = "sglang.cli.main:main"
+
 [tool.setuptools.package-data]
 "sglang" = [
   "srt/**/*",
diff --git a/python/pyproject_xpu.toml b/python/pyproject_xpu.toml
index 0f9dc3466..c9c56e1c2 100644
--- a/python/pyproject_xpu.toml
+++ b/python/pyproject_xpu.toml
@@ -97,6 +97,9 @@ dev = ["sglang[test]"]
 "Homepage" = "https://github.com/sgl-project/sglang"
 "Bug Tracker" = "https://github.com/sgl-project/sglang/issues"
 
+[project.scripts]
+sglang = "sglang.cli.main:main"
+
 [tool.setuptools.package-data]
 "sglang" = [
   "srt/**/*",
diff --git a/python/sglang/srt/server_args.py b/python/sglang/srt/server_args.py
index 07630c9f1..0b8c1b280 100644
--- a/python/sglang/srt/server_args.py
+++ b/python/sglang/srt/server_args.py
@@ -46,6 +46,7 @@ from sglang.srt.utils.common import (
     get_int_env_var,
     get_quantization_config,
     is_blackwell_supported,
+    is_cpu,
     is_cuda,
     is_flashinfer_available,
     is_hip,
@@ -926,8 +927,8 @@ class ServerArgs:
         # 5. Pipeline parallelism
         if self.pp_size > 1:
             self.disable_piecewise_cuda_graph = True
-        # 6. Non-CUDA hardware (AMD, NPU, etc.)
-        if is_hip() or is_npu():
+        # 6. Non-CUDA hardware (AMD, NPU, CPU, etc.)
+        if is_hip() or is_npu() or is_cpu():
             self.disable_piecewise_cuda_graph = True
         # 7. MoE A2A backend
         if self.moe_a2a_backend != "none":