diff --git a/wide_n_deep_serving_client/wide_n_deep_serving_client.cpp b/wide_n_deep_serving_client/wide_n_deep_serving_client.cpp
new file mode 100644
index 0000000..e6d8343
--- /dev/null
+++ b/wide_n_deep_serving_client/wide_n_deep_serving_client.cpp
@@ -0,0 +1,112 @@
+/*
+ * wide_n_deep_serving_client.cpp
+ *
+ *  Created on: 2017年10月28日
+ *      Author: lambdaji
+ */
+
+#include "wide_n_deep_serving_client.h"
+#include "google/protobuf/map.h"
+#include "tensorflow/core/example/example.pb.h"
+#include "tensorflow/core/example/feature.pb.h"
+#include "tensorflow/core/framework/tensor.h"
+#include "tensorflow/core/platform/types.h"
+#include "tensorflow/core/util/command_line_flags.h"
+#include <vector>
+#include <string>
+
+typedef google::protobuf::Map<tensorflow::string, tensorflow::TensorProto> OutMap;
+
+using grpc::Channel;
+using grpc::ClientContext;
+using grpc::Status;
+
+using tensorflow::serving::PredictRequest;
+using tensorflow::serving::PredictResponse;
+
+int ServingClient::callPredict(const std::string& model_name, const std::string& model_signature_name, std::map<std::string, std::string> & result)
+{
+    PredictRequest predictRequest;
+    PredictResponse response;
+    ClientContext context;
+    tensorflow::TensorProto req_tp;
+    tensorflow::Example example;
+    //int64_t iBegin = TNOWMS;
+    //int64_t iEnd   = TNOWMS;
+
+    predictRequest.mutable_model_spec()->set_name(model_name);
+    predictRequest.mutable_model_spec()->set_signature_name(model_signature_name); //serving_default
+
+    //iBegin = TNOWMS;
+    google::protobuf::Map<tensorflow::string, tensorflow::TensorProto>& inputs = *predictRequest.mutable_inputs();
+    google::protobuf::Map<tensorflow::string, tensorflow::Feature>& feature_dict = *example.mutable_features()->mutable_feature();
+
+    //feature to example
+    feature_dict["age"].mutable_float_list()->add_value(25);
+    feature_dict["capital_gain"].mutable_float_list()->add_value(0);
+    feature_dict["capital_loss"].mutable_float_list()->add_value(0);
+    feature_dict["education"].mutable_bytes_list()->add_value("11th");
+    feature_dict["education_num"].mutable_float_list()->add_value(7);
+    feature_dict["gender"].mutable_bytes_list()->add_value("Male");
+    feature_dict["hours_per_week"].mutable_float_list()->add_value(40);
+    feature_dict["native_country"].mutable_bytes_list()->add_value("United-States");
+    feature_dict["occupation"].mutable_bytes_list()->add_value("Machine-op-inspct");
+    feature_dict["relationship"].mutable_bytes_list()->add_value("Own-child");
+    feature_dict["workclass"].mutable_bytes_list()->add_value("Private");
+
+    //serialize to req.inputs
+    string serialized;
+    example.SerializeToString(&serialized);
+    req_tp.mutable_tensor_shape()->add_dim()->set_size(1);		//set_size(5) for batch predicting
+    req_tp.set_dtype(tensorflow::DataType::DT_STRING);
+    req_tp.add_string_val(serialized);		//1st
+    //req_tp.add_string_val(serialized);	//2nd
+    //req_tp.add_string_val(serialized);	//3rd
+    //req_tp.add_string_val(serialized);	//4th
+    //req_tp.add_string_val(serialized);	//5th
+    inputs["inputs"] = req_tp;
+
+    //iEnd   = TNOWMS;
+    //TLOGDEBUG("ServingClient::callPredict sample_to_tfrequest timecost(ms):" << (iEnd - iBegin) << endl);
+
+    //predict
+    //iBegin = TNOWMS;
+    Status status = _stub->Predict(&context, predictRequest, &response);
+    //iEnd   = TNOWMS;
+    //TLOGDEBUG("ServingClient::callPredict _stub->Predict timecost(ms):" << (iEnd - iBegin) << endl);
+
+    if (status.ok())
+    {
+        //TLOGDEBUG("ServingClient::callPredict call predict ok" << endl);
+        //TLOGDEBUG("ServingClient::callPredict outputs size is " << response.outputs_size() << endl);
+
+        OutMap& map_outputs = *response.mutable_outputs();
+        OutMap::iterator iter;
+        int output_index = 0;
+
+        for (iter = map_outputs.begin(); iter != map_outputs.end(); ++iter)
+        {
+            tensorflow::TensorProto& result_tensor_proto = iter->second;
+            tensorflow::Tensor tensor;
+            bool converted = tensor.FromProto(result_tensor_proto);
+            if (converted)
+            {
+                //TLOGDEBUG("ServingClient::callPredict the result tensor[" << output_index << "] is:" << tensor.SummarizeValue(10) << endl);
+                result[iter->first] = tensor.SummarizeValue(10);
+            }
+            else
+            {
+                //TLOGDEBUG("ServingClient::callPredict the result tensor[" << output_index << "] convert failed." << endl);
+            }
+            ++output_index;
+        }
+
+        return 0;
+    }
+    else
+    {
+        //TLOGDEBUG("ServingClient::callPredict gRPC call return code: " << status.error_code() << ": " << status.error_message() << endl);
+        return -1;
+    }
+}
+
diff --git a/wide_n_deep_serving_client/wide_n_deep_serving_client.h b/wide_n_deep_serving_client/wide_n_deep_serving_client.h
new file mode 100644
index 0000000..6f8078c
--- /dev/null
+++ b/wide_n_deep_serving_client/wide_n_deep_serving_client.h
@@ -0,0 +1,41 @@
+/*
+ * wide_n_deep_serving_client.h
+ *
+ *  Created on: 2017年10月28日
+ *      Author: lambdaji
+ */
+
+#ifndef WIDE_N_DEEP_SERVING_CLIENT_H_
+#define WIDE_N_DEEP_SERVING_CLIENT_H_
+
+#include <fstream>
+#include <iostream>
+#include <memory>
+#include <string>
+#include <map>
+#include <vector>
+
+#include "grpc++/create_channel.h"
+#include "grpc++/security/credentials.h"
+#include "tensorflow_serving/apis/prediction_service.grpc.pb.h"
+
+using tensorflow::serving::PredictionService;
+
+class ServingClient
+{
+public:
+    static std::shared_ptr<ServingClient> createClient(const std::string sServerPort){
+        std::shared_ptr<ServingClient> p = std::make_shared<ServingClient>(grpc::CreateChannel(sServerPort, grpc::InsecureChannelCredentials()));
+        return p;
+    }
+public:
+    ServingClient(const std::shared_ptr<grpc::Channel>& channel) : _stub(PredictionService::NewStub(channel)) { }
+    int callPredict(const std::string& model_name, const std::string& model_signature_name, std::map<std::string, std::string> & result);
+
+private:
+    std::unique_ptr<PredictionService::Stub>    _stub;
+};
+
+
+
+#endif /* WIDE_N_DEEP_SERVING_CLIENT_H_ */
diff --git a/wide_n_deep_serving_client/wide_n_deep_serving_client.md b/wide_n_deep_serving_client/wide_n_deep_serving_client.md
new file mode 100644
index 0000000..a8fed30
--- /dev/null
+++ b/wide_n_deep_serving_client/wide_n_deep_serving_client.md
@@ -0,0 +1,12 @@
+Serving a TensorFlow Wide & Deep Model (tf.estimator) by TF-Serving in C++
+
+This tutorial builds on the code developed in TensorFlow Wide & Deep Learning Tutorial, so if you haven't yet completed that tutorial, you need to take a look at it first.
+
+This tutorial referenced a python version from https://github.com/MtDersvan/tf_playground/tree/master/wide_and_deep_tutorial.
+
+You'd better take a look at the following proto file first in:
+ - tensorflow/core/example/example.proto
+ - tensorflow/core/example/feature.proto
+ - tensorflow/core/framework/tensor.proto
+ - tensorflow_serving/apis/predict.proto
+ - tensorflow_serving/apis/model.proto
\ No newline at end of file