diff --git a/src/Makefile b/src/Makefile
index 49bf50f7..74fbc051 100644
--- a/src/Makefile
+++ b/src/Makefile
@@ -357,15 +357,11 @@ ifeq ($(OS), Android)
 endif
 
 ### 3.10 MPI
-ifeq ($(CXX),$(filter $(CXX),mpicxx mpic++ mpiCC mpicxx.mpich))
+ifneq (,$(findstring mpi, $(CXX)))
 	mpi = yes
-endif
-
-ifeq ($(mpi),yes)
 	CXXFLAGS += -DUSE_MPI -Wno-cast-qual
 endif
 
-
 ### ==========================================================================
 ### Section 4. Public targets
 ### ==========================================================================
diff --git a/src/cluster.cpp b/src/cluster.cpp
index ff492b27..e0d3afc9 100644
--- a/src/cluster.cpp
+++ b/src/cluster.cpp
@@ -50,6 +50,7 @@ static std::vector<KeyedTTEntry> TTBuff;
 static MPI_Datatype MIDatatype = MPI_DATATYPE_NULL;
 
 void init() {
+
   int thread_support;
   MPI_Init_thread(nullptr, nullptr, MPI_THREAD_MULTIPLE, &thread_support);
   if (thread_support < MPI_THREAD_MULTIPLE)
@@ -78,10 +79,12 @@ void init() {
 }
 
 void finalize() {
+
   MPI_Finalize();
 }
 
 bool getline(std::istream& input, std::string& str) {
+
   int size;
   std::vector<char> vec;
   bool state;
@@ -98,27 +101,31 @@ bool getline(std::istream& input, std::string& str) {
   MPI_Ibcast(&size, 1, MPI_INT, 0, InputComm, &reqInput);
   if (is_root())
       MPI_Wait(&reqInput, MPI_STATUS_IGNORE);
-  else {
-      while (true) {
+  else
+  {
+      while (true)
+      {
           int flag;
           MPI_Test(&reqInput, &flag, MPI_STATUS_IGNORE);
           if (flag)
               break;
-          else {
+          else
               std::this_thread::sleep_for(std::chrono::milliseconds(10));
-          }
       }
   }
+
   if (!is_root())
       vec.resize(size);
   MPI_Bcast(vec.data(), size, MPI_CHAR, 0, InputComm);
   if (!is_root())
       str.assign(vec.begin(), vec.end());
   MPI_Bcast(&state, 1, MPI_CXX_BOOL, 0, InputComm);
+
   return state;
 }
 
 void sync_start() {
+
   stop_signal = false;
 
   // Start listening to stop signal
@@ -127,15 +134,19 @@ void sync_start() {
 }
 
 void sync_stop() {
-  if (is_root()) {
-      if (!stop_signal && Threads.stop) {
+
+  if (is_root())
+  {
+      if (!stop_signal && Threads.stop)
+      {
           // Signal the cluster about stopping
           stop_signal = true;
           MPI_Ibarrier(StopComm, &reqStop);
           MPI_Wait(&reqStop, MPI_STATUS_IGNORE);
       }
   }
-  else {
+  else
+  {
       int flagStop;
       // Check if we've received any stop signal
       MPI_Test(&reqStop, &flagStop, MPI_STATUS_IGNORE);
@@ -145,10 +156,12 @@ void sync_stop() {
 }
 
 int size() {
+
   return world_size;
 }
 
 int rank() {
+
   return world_rank;
 }
 
@@ -185,6 +198,7 @@ void save(Thread* thread, TTEntry* tte,
              {
                  if (irank == size_t(rank()))
                      continue;
+
                  for (size_t i = irank * TTSendBufferSize ; i < (irank + 1) * TTSendBufferSize; ++i)
                  {
                      auto&& e = TTBuff[i];
@@ -198,7 +212,8 @@ void save(Thread* thread, TTEntry* tte,
              send_buff = {};
 
              // Build up new send buffer: best 16 found across all threads
-             for (auto&& th : Threads) {
+             for (auto&& th : Threads)
+             {
                  std::lock_guard<Mutex> lk(th->ttBuffer.mutex);
                  for (auto&& e : th->ttBuffer.buffer)
                      send_buff.replace(e);
@@ -216,24 +231,32 @@ void save(Thread* thread, TTEntry* tte,
 }
 
 void pick_moves(MoveInfo& mi) {
+
   MoveInfo* pMoveInfo = NULL;
-  if (is_root()) {
+  if (is_root())
+  {
       pMoveInfo = (MoveInfo*)malloc(sizeof(MoveInfo) * size());
   }
   MPI_Gather(&mi, 1, MIDatatype, pMoveInfo, 1, MIDatatype, 0, MoveComm);
-  if (is_root()) {
+
+  if (is_root())
+  {
       std::map<int, int> votes;
       int minScore = pMoveInfo[0].score;
-      for (int i = 0; i < size(); i++) {
+      for (int i = 0; i < size(); ++i)
+      {
           minScore = std::min(minScore, pMoveInfo[i].score);
           votes[pMoveInfo[i].move] = 0;
       }
-      for (int i = 0; i < size(); i++) {
+      for (int i = 0; i < size(); ++i)
+      {
           votes[pMoveInfo[i].move] += pMoveInfo[i].score - minScore + pMoveInfo[i].depth;
       }
       int bestVote = votes[pMoveInfo[0].move];
-      for (int i = 0; i < size(); i++) {
-          if (votes[pMoveInfo[i].move] > bestVote) {
+      for (int i = 0; i < size(); ++i)
+      {
+          if (votes[pMoveInfo[i].move] > bestVote)
+          {
               bestVote = votes[pMoveInfo[i].move];
               mi = pMoveInfo[i];
           }
diff --git a/src/cluster.h b/src/cluster.h
index aea6a12e..3bbe9b9c 100644
--- a/src/cluster.h
+++ b/src/cluster.h
@@ -44,6 +44,7 @@ using KeyedTTEntry = std::pair<Key, TTEntry>;
 
 constexpr std::size_t TTSendBufferSize = 16;
 template <std::size_t N> class TTSendBuffer : public std::array<KeyedTTEntry, N> {
+
   struct Compare {
       inline bool operator()(const KeyedTTEntry& lhs, const KeyedTTEntry& rhs) {
           return lhs.second.depth() > rhs.second.depth();
@@ -53,7 +54,9 @@ template <std::size_t N> class TTSendBuffer : public std::array<KeyedTTEntry, N>
 
 public:
   bool replace(const KeyedTTEntry& value) {
-      if (compare(value, this->front())) {
+
+      if (compare(value, this->front()))
+      {
           std::pop_heap(this->begin(), this->end(), compare);
           this->back() = value;
           std::push_heap(this->begin(), this->end(), compare);
@@ -69,8 +72,7 @@ bool getline(std::istream& input, std::string& str);
 int size();
 int rank();
 inline bool is_root() { return rank() == 0; }
-void save(Thread* thread, TTEntry* tte,
-          Key k, Value v, Bound b, Depth d, Move m, Value ev);
+void save(Thread* thread, TTEntry* tte, Key k, Value v, Bound b, Depth d, Move m, Value ev);
 void pick_moves(MoveInfo& mi);
 void sync_start();
 void sync_stop();
@@ -80,13 +82,14 @@ void sync_stop();
 inline void init() { }
 inline void finalize() { }
 inline bool getline(std::istream& input, std::string& str) {
+
   return static_cast<bool>(std::getline(input, str));
 }
 constexpr int size() { return 1; }
 constexpr int rank() { return 0; }
 constexpr bool is_root() { return true; }
-inline void save(Thread* thread, TTEntry* tte,
-          Key k, Value v, Bound b, Depth d, Move m, Value ev) {
+inline void save(Thread* thread, TTEntry* tte, Key k, Value v, Bound b, Depth d, Move m, Value ev) {
+
   (void)thread;
   tte->save(k, v, b, d, m, ev);
 }
diff --git a/src/search.cpp b/src/search.cpp
index 89214e22..0956180b 100644
--- a/src/search.cpp
+++ b/src/search.cpp
@@ -296,6 +296,7 @@ void MainThread::search() {
 
   previousScore = static_cast<Value>(mi.score);
 
+  // TODO output should be done on the cluster_root
   if (mi.rank == Cluster::rank()) {
       // Send again PV info if we have a new best thread
       if (!Cluster::is_root() || bestThread != this)
diff --git a/src/uci.cpp b/src/uci.cpp
index 00b3d10d..a62edce7 100644
--- a/src/uci.cpp
+++ b/src/uci.cpp
@@ -141,6 +141,8 @@ namespace {
   // a list of UCI commands is setup according to bench parameters, then
   // it is run one by one printing a summary at the end.
 
+  // TODO make (output?) cluster compatible
+
   void bench(Position& pos, istream& args, StateListPtr& states) {
 
     string token;