diff --git a/vllm/v1/worker/cp_utils.py b/vllm/v1/worker/cp_utils.py
index f666c739b0be7..7625b1b5f8951 100644
--- a/vllm/v1/worker/cp_utils.py
+++ b/vllm/v1/worker/cp_utils.py
@@ -21,22 +21,58 @@ def check_attention_cp_compatibility(vllm_config: VllmConfig) -> None:
             layer_impl = getattr(layer, "impl", None)
             if layer_impl is None:
                 continue
-            if vllm_config.speculative_config is not None and interleave_size > 1:
-                assert layer_impl.supports_mtp_with_cp_non_trivial_interleave_size, (
-                    "MTP with cp_kv_cache_interleave_size > 1 is not "
-                    f"supported in {layer_impl.__class__.__name__}."
-                )
-            if dcp_size > 1:
-                assert layer_impl.need_to_return_lse_for_decode, (
-                    "DCP requires attention impls to return"
-                    " the softmax lse for decode, but the impl "
-                    f"{layer_impl.__class__.__name__} "
-                    "does not return the softmax lse for decode."
+
+            supports_mtp = layer_impl.supports_mtp_with_cp_non_trivial_interleave_size
+            if (
+                vllm_config.speculative_config is not None
+                and interleave_size > 1
+                and not supports_mtp
+            ):
+                raise RuntimeError(
+                    f"Multi-Token Prediction (MTP) with "
+                    f"cp_kv_cache_interleave_size > 1 is not supported by the "
+                    f"current attention backend "
+                    f"'{layer_impl.__class__.__name__}'.\n\n"
+                    f"To resolve this issue, try one of the following:\n"
+                    f"  1. Set cp_kv_cache_interleave_size to 1\n"
+                    f"  2. Disable speculative decoding\n\n"
+                    f"Note: No backends currently support MTP with "
+                    f"cp_kv_cache_interleave_size > 1.\n\n"
+                    f"For more information, see:\n"
+                    f"  https://docs.vllm.ai/en/latest/serving/"
+                    f"distributed_serving.html"
                 )
 
-            if pcp_size > 1:
-                assert layer_impl.supports_pcp, (
-                    "PCP requires attention impls' support, "
-                    f"but the impl {layer_impl.__class__.__name__} "
-                    "does not support PCP."
+            if dcp_size > 1 and not layer_impl.need_to_return_lse_for_decode:
+                raise RuntimeError(
+                    f"Decode Context Parallel (DCP) requires an attention "
+                    f"backend that supports returning softmax LSE (log-sum-exp) "
+                    f"for decode operations. The current backend "
+                    f"'{layer_impl.__class__.__name__}' does not support this "
+                    f"feature.\n\n"
+                    f"To resolve this issue, try one of the following:\n"
+                    f"  1. Use a compatible attention backend by specifying:\n"
+                    f"     --attention-backend <backend>\n"
+                    f"     Compatible backends: FLASH_ATTN, FLASHINFER, "
+                    f"TRITON_MLA, FLASH_MLA, FLASH_ATTN_MLA, CUTLASS_MLA\n"
+                    f"  2. Disable DCP by removing the "
+                    f"--decode-context-parallel-size flag\n\n"
+                    f"For more information, see:\n"
+                    f"  https://docs.vllm.ai/en/latest/serving/"
+                    f"distributed_serving.html"
+                )
+
+            if pcp_size > 1 and not layer_impl.supports_pcp:
+                raise RuntimeError(
+                    f"Prefill Context Parallel (PCP) requires an attention "
+                    f"backend that supports PCP. The current backend "
+                    f"'{layer_impl.__class__.__name__}' does not support this "
+                    f"feature.\n\n"
+                    f"To resolve this issue:\n"
+                    f"  Disable PCP by removing the "
+                    f"--prefill-context-parallel-size flag\n\n"
+                    f"Note: No backends currently support PCP.\n\n"
+                    f"For more information, see:\n"
+                    f"  https://docs.vllm.ai/en/latest/serving/"
+                    f"distributed_serving.html"
                 )