diff --git a/lib/mammoth/include/mammoth/endpoint_server.h b/lib/mammoth/include/mammoth/endpoint_server.h
index c4e3e5a..7224d11 100644
--- a/lib/mammoth/include/mammoth/endpoint_server.h
+++ b/lib/mammoth/include/mammoth/endpoint_server.h
@@ -5,6 +5,8 @@
 #include <ztypes.h>
 
 #include "mammoth/endpoint_client.h"
+#include "mammoth/request_context.h"
+#include "mammoth/response_context.h"
 
 class EndpointServer {
  public:
@@ -12,19 +14,24 @@ class EndpointServer {
   EndpointServer(const EndpointServer&) = delete;
   EndpointServer& operator=(const EndpointServer&) = delete;
 
-  static glcr::ErrorOr<glcr::UniquePtr<EndpointServer>> Create();
-  static glcr::UniquePtr<EndpointServer> Adopt(z_cap_t endpoint_cap);
-
   glcr::ErrorOr<glcr::UniquePtr<EndpointClient>> CreateClient();
-
   // FIXME: Release Cap here.
   z_cap_t GetCap() { return endpoint_cap_; }
 
-  glcr::ErrorCode Recieve(uint64_t* num_bytes, void* data,
+  glcr::ErrorCode Receive(uint64_t* num_bytes, void* data,
                           z_cap_t* reply_port_cap);
 
+  glcr::ErrorCode RunServer();
+
+  virtual glcr::ErrorCode HandleRequest(RequestContext& request,
+                                        ResponseContext& response) = 0;
+
+ protected:
+  EndpointServer(z_cap_t cap) : endpoint_cap_(cap) {}
+
  private:
   z_cap_t endpoint_cap_;
 
-  EndpointServer(z_cap_t cap) : endpoint_cap_(cap) {}
+  static const uint64_t kBufferSize = 1024;
+  uint8_t recieve_buffer_[kBufferSize];
 };
diff --git a/lib/mammoth/include/mammoth/request_context.h b/lib/mammoth/include/mammoth/request_context.h
new file mode 100644
index 0000000..0f8ca96
--- /dev/null
+++ b/lib/mammoth/include/mammoth/request_context.h
@@ -0,0 +1,32 @@
+#pragma once
+
+#include <glacier/status/error.h>
+#include <stdint.h>
+
+class RequestContext {
+ public:
+  RequestContext(void* buffer, uint64_t buffer_length)
+      : buffer_(buffer), buffer_length_(buffer_length) {
+    if (buffer_length_ < sizeof(uint64_t)) {
+      request_id_ = -1;
+    } else {
+      request_id_ = *reinterpret_cast<uint64_t*>(buffer);
+    }
+  }
+
+  uint64_t request_id() { return request_id_; }
+
+  template <typename T>
+  glcr::ErrorCode As(T** arg) {
+    if (buffer_length_ < sizeof(T)) {
+      return glcr::INVALID_ARGUMENT;
+    }
+    *arg = reinterpret_cast<T*>(buffer_);
+    return glcr::OK;
+  }
+
+ private:
+  uint64_t request_id_;
+  void* buffer_;
+  uint64_t buffer_length_;
+};
diff --git a/lib/mammoth/include/mammoth/response_context.h b/lib/mammoth/include/mammoth/response_context.h
new file mode 100644
index 0000000..fbd0d2a
--- /dev/null
+++ b/lib/mammoth/include/mammoth/response_context.h
@@ -0,0 +1,40 @@
+#pragma once
+
+#include <glacier/status/error.h>
+#include <zcall.h>
+#include <ztypes.h>
+
+class ResponseContext {
+ public:
+  ResponseContext(z_cap_t reply_port) : reply_port_(reply_port) {}
+
+  ResponseContext(ResponseContext&) = delete;
+
+  template <typename T>
+  glcr::ErrorCode WriteStruct(const T& response) {
+    // FIXME: Here and below probably don't count as written on error.
+    written_ = true;
+    return ZReplyPortSend(reply_port_, sizeof(T), &response, 0, nullptr);
+  }
+
+  template <typename T>
+  glcr::ErrorCode WriteStructWithCap(const T& response, z_cap_t capability) {
+    written_ = true;
+    return ZReplyPortSend(reply_port_, sizeof(T), &response, 1, &capability);
+  }
+
+  glcr::ErrorCode WriteError(glcr::ErrorCode code) {
+    uint64_t response[2]{
+        static_cast<uint64_t>(-1),
+        code,
+    };
+    written_ = true;
+    return ZReplyPortSend(reply_port_, sizeof(response), &response, 0, nullptr);
+  }
+
+  bool HasWritten() { return written_; }
+
+ private:
+  z_cap_t reply_port_;
+  bool written_ = false;
+};
diff --git a/lib/mammoth/src/endpoint_server.cpp b/lib/mammoth/src/endpoint_server.cpp
index cacc5a6..42f5ecd 100644
--- a/lib/mammoth/src/endpoint_server.cpp
+++ b/lib/mammoth/src/endpoint_server.cpp
@@ -1,14 +1,6 @@
 #include "mammoth/endpoint_server.h"
 
-glcr::ErrorOr<glcr::UniquePtr<EndpointServer>> EndpointServer::Create() {
-  uint64_t cap;
-  RET_ERR(ZEndpointCreate(&cap));
-  return glcr::UniquePtr<EndpointServer>(new EndpointServer(cap));
-}
-
-glcr::UniquePtr<EndpointServer> EndpointServer::Adopt(z_cap_t endpoint_cap) {
-  return glcr::UniquePtr<EndpointServer>(new EndpointServer(endpoint_cap));
-}
+#include "mammoth/debug.h"
 
 glcr::ErrorOr<glcr::UniquePtr<EndpointClient>> EndpointServer::CreateClient() {
   uint64_t client_cap;
@@ -17,7 +9,24 @@ glcr::ErrorOr<glcr::UniquePtr<EndpointClient>> EndpointServer::CreateClient() {
   return EndpointClient::AdoptEndpoint(client_cap);
 }
 
-glcr::ErrorCode EndpointServer::Recieve(uint64_t* num_bytes, void* data,
+glcr::ErrorCode EndpointServer::Receive(uint64_t* num_bytes, void* data,
                                         z_cap_t* reply_port_cap) {
   return ZEndpointRecv(endpoint_cap_, num_bytes, data, reply_port_cap);
 }
+
+glcr::ErrorCode EndpointServer::RunServer() {
+  while (true) {
+    uint64_t message_size = kBufferSize;
+    uint64_t reply_port_cap = 0;
+    RET_ERR(Receive(&message_size, recieve_buffer_, &reply_port_cap));
+
+    RequestContext request(recieve_buffer_, message_size);
+    ResponseContext response(reply_port_cap);
+    // FIXME: Consider pumping these errors into the response as well.
+    RET_ERR(HandleRequest(request, response));
+    if (!response.HasWritten()) {
+      dbgln("Returning without having written a response. Req type %x",
+            request.request_id());
+    }
+  }
+}
diff --git a/sys/denali/ahci/command.cpp b/sys/denali/ahci/command.cpp
index aa8a020..986ca6d 100644
--- a/sys/denali/ahci/command.cpp
+++ b/sys/denali/ahci/command.cpp
@@ -7,8 +7,8 @@
 Command::~Command() {}
 
 DmaReadCommand::DmaReadCommand(uint64_t lba, uint64_t sector_cnt,
-                               DmaCallback callback, z_cap_t reply_port)
-    : reply_port_(reply_port),
+                               DmaCallback callback, ResponseContext& response)
+    : response_(response),
       lba_(lba),
       sector_cnt_(sector_cnt),
       callback_(callback) {
@@ -50,5 +50,5 @@ void DmaReadCommand::PopulatePrdt(PhysicalRegionDescriptor* prdt) {
   prdt[0].byte_count = region_.size();
 }
 void DmaReadCommand::Callback() {
-  callback_(reply_port_, lba_, sector_cnt_, region_.cap());
+  callback_(response_, lba_, sector_cnt_, region_.cap());
 }
diff --git a/sys/denali/ahci/command.h b/sys/denali/ahci/command.h
index f502b01..9870685 100644
--- a/sys/denali/ahci/command.h
+++ b/sys/denali/ahci/command.h
@@ -1,6 +1,7 @@
 #pragma once
 
 #include <mammoth/memory_region.h>
+#include <mammoth/response_context.h>
 #include <stdint.h>
 
 #include "ahci/ahci.h"
@@ -15,9 +16,9 @@ class Command {
 
 class DmaReadCommand : public Command {
  public:
-  typedef void (*DmaCallback)(z_cap_t, uint64_t, uint64_t, z_cap_t);
+  typedef void (*DmaCallback)(ResponseContext&, uint64_t, uint64_t, z_cap_t);
   DmaReadCommand(uint64_t lba, uint64_t sector_cnt, DmaCallback callback,
-                 z_cap_t reply_port);
+                 ResponseContext& reply_port);
 
   virtual ~DmaReadCommand() override;
 
@@ -27,7 +28,7 @@ class DmaReadCommand : public Command {
   void Callback() override;
 
  private:
-  z_cap_t reply_port_;
+  ResponseContext& response_;
   uint64_t lba_;
   uint64_t sector_cnt_;
   DmaCallback callback_;
diff --git a/sys/denali/client/denali_client.cpp b/sys/denali/client/denali_client.cpp
index 6f870b2..9d95f00 100644
--- a/sys/denali/client/denali_client.cpp
+++ b/sys/denali/client/denali_client.cpp
@@ -6,13 +6,14 @@
 
 glcr::ErrorOr<MappedMemoryRegion> DenaliClient::ReadSectors(
     uint64_t device_id, uint64_t lba, uint64_t num_sectors) {
-  DenaliRead read{
+  DenaliReadRequest read{
       .device_id = device_id,
       .lba = lba,
       .size = num_sectors,
   };
   auto pair_or =
-      endpoint_->CallEndpointGetCap<DenaliRead, DenaliReadResponse>(read);
+      endpoint_->CallEndpointGetCap<DenaliReadRequest, DenaliReadResponse>(
+          read);
   if (!pair_or) {
     return pair_or.error();
   }
diff --git a/sys/denali/denali.cpp b/sys/denali/denali.cpp
index 382ec07..043de69 100644
--- a/sys/denali/denali.cpp
+++ b/sys/denali/denali.cpp
@@ -19,14 +19,14 @@ uint64_t main(uint64_t init_port_cap) {
   ASSIGN_OR_RETURN(MappedMemoryRegion ahci_region, stub.GetAhciConfig());
   ASSIGN_OR_RETURN(auto driver, AhciDriver::Init(ahci_region));
 
-  ASSIGN_OR_RETURN(glcr::UniquePtr<EndpointServer> endpoint,
-                   EndpointServer::Create());
+  ASSIGN_OR_RETURN(glcr::UniquePtr<DenaliServer> server,
+                   DenaliServer::Create(*driver));
+
   ASSIGN_OR_RETURN(glcr::UniquePtr<EndpointClient> client,
-                   endpoint->CreateClient());
+                   server->CreateClient());
   check(stub.Register("denali", *client));
 
-  DenaliServer server(glcr::Move(endpoint), *driver);
-  RET_ERR(server.RunServer());
+  RET_ERR(server->RunServer());
   // FIXME: Add thread join.
   return 0;
 }
diff --git a/sys/denali/denali_server.cpp b/sys/denali/denali_server.cpp
index f34b21a..caa29a4 100644
--- a/sys/denali/denali_server.cpp
+++ b/sys/denali/denali_server.cpp
@@ -7,61 +7,57 @@
 
 namespace {
 DenaliServer* gServer = nullptr;
-void HandleResponse(z_cap_t reply_port, uint64_t lba, uint64_t size,
+void HandleResponse(ResponseContext& response, uint64_t lba, uint64_t size,
                     z_cap_t mem) {
-  gServer->HandleResponse(reply_port, lba, size, mem);
+  gServer->HandleResponse(response, lba, size, mem);
 }
 }  // namespace
 
-DenaliServer::DenaliServer(glcr::UniquePtr<EndpointServer> server,
-                           AhciDriver& driver)
-    : server_(glcr::Move(server)), driver_(driver) {
-  gServer = this;
+glcr::ErrorOr<glcr::UniquePtr<DenaliServer>> DenaliServer::Create(
+    AhciDriver& driver) {
+  z_cap_t cap;
+  RET_ERR(ZEndpointCreate(&cap));
+  return glcr::UniquePtr<DenaliServer>(new DenaliServer(cap, driver));
 }
 
-glcr::ErrorCode DenaliServer::RunServer() {
-  while (true) {
-    uint64_t buff_size = kBuffSize;
-    z_cap_t reply_port;
-    RET_ERR(server_->Recieve(&buff_size, read_buffer_, &reply_port));
-    if (buff_size < sizeof(uint64_t)) {
-      dbgln("Skipping invalid message");
-      continue;
-    }
-    uint64_t type = *reinterpret_cast<uint64_t*>(read_buffer_);
-    switch (type) {
-      case Z_INVALID:
-        dbgln(reinterpret_cast<char*>(read_buffer_));
-        break;
-      case DENALI_READ: {
-        DenaliRead* read_req = reinterpret_cast<DenaliRead*>(read_buffer_);
-        uint64_t memcap = 0;
-        RET_ERR(HandleRead(*read_req, reply_port));
-        break;
+glcr::ErrorCode DenaliServer::HandleRequest(RequestContext& request,
+                                            ResponseContext& response) {
+  switch (request.request_id()) {
+    case DENALI_READ: {
+      DenaliReadRequest* req = 0;
+      glcr::ErrorCode err = request.As<DenaliReadRequest>(&req);
+      if (err != glcr::OK) {
+        response.WriteError(err);
       }
-      default:
-        dbgln("Invalid message type.");
-        return glcr::UNIMPLEMENTED;
+      err = HandleRead(req, response);
+      if (err != glcr::OK) {
+        response.WriteError(err);
+      }
+      break;
     }
+    default:
+      response.WriteError(glcr::UNIMPLEMENTED);
+      break;
   }
+  return glcr::OK;
 }
 
-glcr::ErrorCode DenaliServer::HandleRead(const DenaliRead& read,
-                                         z_cap_t reply_port) {
-  ASSIGN_OR_RETURN(AhciDevice * device, driver_.GetDevice(read.device_id));
+glcr::ErrorCode DenaliServer::HandleRead(DenaliReadRequest* request,
+                                         ResponseContext& context) {
+  ASSIGN_OR_RETURN(AhciDevice * device, driver_.GetDevice(request->device_id));
 
-  device->IssueCommand(
-      new DmaReadCommand(read.lba, read.size, ::HandleResponse, reply_port));
+  device->IssueCommand(new DmaReadCommand(request->lba, request->size,
+                                          ::HandleResponse, context));
 
   return glcr::OK;
 }
 
-void DenaliServer::HandleResponse(z_cap_t reply_port, uint64_t lba,
+void DenaliServer::HandleResponse(ResponseContext& response, uint64_t lba,
                                   uint64_t size, z_cap_t mem) {
   DenaliReadResponse resp{
       .device_id = 0,
       .lba = lba,
       .size = size,
   };
-  check(ZReplyPortSend(reply_port, sizeof(resp), &resp, 1, &mem));
+  check(response.WriteStructWithCap<DenaliReadResponse>(resp, mem));
 }
diff --git a/sys/denali/denali_server.h b/sys/denali/denali_server.h
index 2cef169..677185d 100644
--- a/sys/denali/denali_server.h
+++ b/sys/denali/denali_server.h
@@ -6,21 +6,26 @@
 #include "ahci/ahci_driver.h"
 #include "denali/denali.h"
 
-class DenaliServer {
+class DenaliServer : public EndpointServer {
  public:
-  DenaliServer(glcr::UniquePtr<EndpointServer> server, AhciDriver& driver);
+  static glcr::ErrorOr<glcr::UniquePtr<DenaliServer>> Create(
+      AhciDriver& driver);
 
-  glcr::ErrorCode RunServer();
-
-  void HandleResponse(z_cap_t reply_port, uint64_t lba, uint64_t size,
+  void HandleResponse(ResponseContext& response, uint64_t lba, uint64_t size,
                       z_cap_t cap);
 
+  virtual glcr::ErrorCode HandleRequest(RequestContext& request,
+                                        ResponseContext& response) override;
+
  private:
   static const uint64_t kBuffSize = 1024;
-  glcr::UniquePtr<EndpointServer> server_;
   uint8_t read_buffer_[kBuffSize];
 
   AhciDriver& driver_;
 
-  glcr::ErrorCode HandleRead(const DenaliRead& read, z_cap_t reply_port);
+  DenaliServer(z_cap_t endpoint_cap, AhciDriver& driver)
+      : EndpointServer(endpoint_cap), driver_(driver) {}
+
+  glcr::ErrorCode HandleRead(DenaliReadRequest* request,
+                             ResponseContext& context);
 };
diff --git a/sys/denali/include/denali/denali.h b/sys/denali/include/denali/denali.h
index 806592a..398e68c 100644
--- a/sys/denali/include/denali/denali.h
+++ b/sys/denali/include/denali/denali.h
@@ -5,7 +5,7 @@
 #define DENALI_INVALID 0
 #define DENALI_READ 100
 
-struct DenaliRead {
+struct DenaliReadRequest {
   uint64_t request_type = DENALI_READ;
   uint64_t device_id;
 
diff --git a/sys/yellowstone/yellowstone.cpp b/sys/yellowstone/yellowstone.cpp
index 6d3a3fe..d7ec505 100644
--- a/sys/yellowstone/yellowstone.cpp
+++ b/sys/yellowstone/yellowstone.cpp
@@ -13,16 +13,15 @@ uint64_t main(uint64_t port_cap) {
   check(ParseInitPort(port_cap));
 
   ASSIGN_OR_RETURN(auto server, YellowstoneServer::Create());
-  Thread server_thread = server->RunServer();
   Thread registration_thread = server->RunRegistration();
 
   uint64_t vaddr;
   check(ZAddressSpaceMap(gSelfVmasCap, 0, gBootDenaliVmmoCap, &vaddr));
   ASSIGN_OR_RETURN(glcr::UniquePtr<EndpointClient> client,
-                   server->GetServerClient());
+                   server->CreateClient());
   check(SpawnProcessFromElfRegion(vaddr, glcr::Move(client)));
 
-  check(server_thread.Join());
+  check(server->RunServer());
   check(registration_thread.Join());
   dbgln("Yellowstone Finished Successfully.");
   return 0;
diff --git a/sys/yellowstone/yellowstone_server.cpp b/sys/yellowstone/yellowstone_server.cpp
index 244650c..e1d965f 100644
--- a/sys/yellowstone/yellowstone_server.cpp
+++ b/sys/yellowstone/yellowstone_server.cpp
@@ -12,11 +12,6 @@
 
 namespace {
 
-void ServerThreadBootstrap(void* yellowstone) {
-  dbgln("Yellowstone server starting");
-  static_cast<YellowstoneServer*>(yellowstone)->ServerThread();
-}
-
 void RegistrationThreadBootstrap(void* yellowstone) {
   dbgln("Yellowstone registration starting");
   static_cast<YellowstoneServer*>(yellowstone)->RegistrationThread();
@@ -40,71 +35,60 @@ glcr::ErrorOr<PartitionInfo> HandleDenaliRegistration(z_cap_t endpoint_cap) {
 }  // namespace
 
 glcr::ErrorOr<glcr::UniquePtr<YellowstoneServer>> YellowstoneServer::Create() {
-  ASSIGN_OR_RETURN(auto server, EndpointServer::Create());
+  z_cap_t cap;
+  RET_ERR(ZEndpointCreate(&cap));
   ASSIGN_OR_RETURN(PortServer port, PortServer::Create());
-  return glcr::UniquePtr<YellowstoneServer>(
-      new YellowstoneServer(glcr::Move(server), port));
+  return glcr::UniquePtr<YellowstoneServer>(new YellowstoneServer(cap, port));
 }
 
-YellowstoneServer::YellowstoneServer(glcr::UniquePtr<EndpointServer> server,
-                                     PortServer port)
-    : server_(glcr::Move(server)), register_port_(port) {}
-
-Thread YellowstoneServer::RunServer() {
-  return Thread(ServerThreadBootstrap, this);
-}
+YellowstoneServer::YellowstoneServer(z_cap_t endpoint_cap, PortServer port)
+    : EndpointServer(endpoint_cap), register_port_(port) {}
 
 Thread YellowstoneServer::RunRegistration() {
   return Thread(RegistrationThreadBootstrap, this);
 }
 
-void YellowstoneServer::ServerThread() {
-  while (true) {
-    uint64_t num_bytes = kBufferSize;
-    uint64_t reply_port_cap;
-    // FIXME: Error handling.
-    check(server_->Recieve(&num_bytes, server_buffer_, &reply_port_cap));
-    YellowstoneGetReq* req =
-        reinterpret_cast<YellowstoneGetReq*>(server_buffer_);
-    switch (req->type) {
-      case kYellowstoneGetAhci: {
-        dbgln("Yellowstone::GetAHCI");
-        YellowstoneGetAhciResp resp{
-            .type = kYellowstoneGetAhci,
-            .ahci_phys_offset = pci_reader_.GetAhciPhysical(),
-        };
-        check(ZReplyPortSend(reply_port_cap, sizeof(resp), &resp, 0, nullptr));
-        break;
-      }
-      case kYellowstoneGetRegistration: {
-        dbgln("Yellowstone::GetRegistration");
-        auto client_or = register_port_.CreateClient();
-        if (!client_or.ok()) {
-          check(client_or.error());
-        }
-        YellowstoneGetRegistrationResp resp;
-        uint64_t reg_cap = client_or.value().cap();
-        check(ZReplyPortSend(reply_port_cap, sizeof(resp), &resp, 1, &reg_cap));
-        break;
-      }
-      case kYellowstoneGetDenali: {
-        dbgln("Yellowstone::GetDenali");
-        z_cap_t new_denali;
-        check(ZCapDuplicate(denali_cap_, &new_denali));
-        YellowstoneGetDenaliResp resp{
-            .type = kYellowstoneGetDenali,
-            .device_id = device_id_,
-            .lba_offset = lba_offset_,
-        };
-        check(ZReplyPortSend(reply_port_cap, sizeof(resp), &resp, 1,
-                             &new_denali));
-        break;
-      }
-      default:
-        dbgln("Unknown request type: %x", req->type);
-        break;
+glcr::ErrorCode YellowstoneServer::HandleRequest(RequestContext& request,
+                                                 ResponseContext& response) {
+  switch (request.request_id()) {
+    case kYellowstoneGetAhci: {
+      dbgln("Yellowstone::GetAHCI");
+      YellowstoneGetAhciResp resp{
+          .type = kYellowstoneGetAhci,
+          .ahci_phys_offset = pci_reader_.GetAhciPhysical(),
+      };
+      RET_ERR(response.WriteStruct<YellowstoneGetAhciResp>(resp));
+      break;
     }
+    case kYellowstoneGetRegistration: {
+      dbgln("Yellowstone::GetRegistration");
+      auto client_or = register_port_.CreateClient();
+      if (!client_or.ok()) {
+        check(client_or.error());
+      }
+      YellowstoneGetRegistrationResp resp;
+      uint64_t reg_cap = client_or.value().cap();
+      RET_ERR(response.WriteStructWithCap(resp, reg_cap));
+      break;
+    }
+    case kYellowstoneGetDenali: {
+      dbgln("Yellowstone::GetDenali");
+      z_cap_t new_denali;
+      check(ZCapDuplicate(denali_cap_, &new_denali));
+      YellowstoneGetDenaliResp resp{
+          .type = kYellowstoneGetDenali,
+          .device_id = device_id_,
+          .lba_offset = lba_offset_,
+      };
+      RET_ERR(response.WriteStructWithCap(resp, new_denali));
+      break;
+    }
+    default:
+      dbgln("Unknown request type: %x", request.request_id());
+      return glcr::UNIMPLEMENTED;
+      break;
   }
+  return glcr::OK;
 }
 
 void YellowstoneServer::RegistrationThread() {
@@ -127,7 +111,7 @@ void YellowstoneServer::RegistrationThread() {
       uint64_t vaddr;
       check(
           ZAddressSpaceMap(gSelfVmasCap, 0, gBootVictoriaFallsVmmoCap, &vaddr));
-      auto client_or = GetServerClient();
+      auto client_or = CreateClient();
       if (!client_or.ok()) {
         check(client_or.error());
       }
@@ -144,8 +128,3 @@ void YellowstoneServer::RegistrationThread() {
     dbgln(name.cstr());
   }
 }
-
-glcr::ErrorOr<glcr::UniquePtr<EndpointClient>>
-YellowstoneServer::GetServerClient() {
-  return server_->CreateClient();
-}
diff --git a/sys/yellowstone/yellowstone_server.h b/sys/yellowstone/yellowstone_server.h
index 609f238..bd643dc 100644
--- a/sys/yellowstone/yellowstone_server.h
+++ b/sys/yellowstone/yellowstone_server.h
@@ -8,20 +8,19 @@
 
 #include "hw/pcie.h"
 
-class YellowstoneServer {
+class YellowstoneServer : public EndpointServer {
  public:
   static glcr::ErrorOr<glcr::UniquePtr<YellowstoneServer>> Create();
 
-  Thread RunServer();
   Thread RunRegistration();
 
-  void ServerThread();
   void RegistrationThread();
 
-  glcr::ErrorOr<glcr::UniquePtr<EndpointClient>> GetServerClient();
+  virtual glcr::ErrorCode HandleRequest(RequestContext& request,
+                                        ResponseContext& response) override;
 
  private:
-  glcr::UniquePtr<EndpointServer> server_;
+  // FIXME: Separate this to its own service.
   PortServer register_port_;
 
   static const uint64_t kBufferSize = 128;
@@ -36,5 +35,5 @@ class YellowstoneServer {
 
   PciReader pci_reader_;
 
-  YellowstoneServer(glcr::UniquePtr<EndpointServer> server, PortServer port);
+  YellowstoneServer(z_cap_t endpoint_cap, PortServer port);
 };