diff --git a/include/margo.h b/include/margo.h
index b4d51f54e28da2de3d19ed46f79191ec6e7919e5..f0229747924f6f358db91f2e985b17eca351567b 100644
--- a/include/margo.h
+++ b/include/margo.h
@@ -100,6 +100,23 @@ void margo_finalize(
 void margo_wait_for_finalize(
     margo_instance_id mid);
 
+/**
+ * Installs a callback to be called before the margo instance is finalize.
+ * Callbacks installed will be called in reverse ordered than they have been
+ * pushed, and with the user-provider pointer as argument.
+ *
+ * Note that callbacks may not be called within margo_finalize. They are called
+ * when the margo instance is cleaned up, which may happen in margo_wait_for_finalize.
+ *
+ * @param mid The margo instance
+ * @param cb Callback to install
+ * @param uargs User-provider argument to pass to the callback when called
+ */
+void margo_push_finalize_callback(
+    margo_instance_id mid,
+    void(*cb)(void*), 
+    void* uargs);
+
 /** 
  * Registers an RPC with margo
  *
diff --git a/src/margo.c b/src/margo.c
index 0ce1b3c490d9e103bb9721f7b4b909ba36024560..36fe8cd54179c0139643cdbe9aa996ef43ad9721 100644
--- a/src/margo.c
+++ b/src/margo.c
@@ -63,6 +63,13 @@ struct margo_handle_cache_el
     struct margo_handle_cache_el *next; /* free list link */
 };
 
+struct margo_finalize_cb
+{
+    void(*callback)(void*);
+    void* uargs;
+    struct margo_finalize_cb* next;
+};
+
 struct margo_instance
 {
     /* mercury/argobots state */
@@ -87,6 +94,7 @@ struct margo_instance
     int refcount;
     ABT_mutex finalize_mutex;
     ABT_cond finalize_cond;
+    struct margo_finalize_cb* finalize_cb;
 
     /* hash table to track multiplexed rpcs registered with margo */
     struct mplex_element *mplex_table;
@@ -279,6 +287,7 @@ margo_instance_id margo_init_pool(ABT_pool progress_pool, ABT_pool handler_pool,
     mid->hg_progress_timeout_ub = DEFAULT_MERCURY_PROGRESS_TIMEOUT_UB;
     mid->mplex_table = NULL;
     mid->refcount = 1;
+    mid->finalize_cb = NULL;
 
     ret = margo_timer_instance_init(mid);
     if(ret != 0) goto err;
@@ -309,6 +318,15 @@ static void margo_cleanup(margo_instance_id mid)
 {
     int i;
 
+    /* call finalize callbacks */
+    struct margo_finalize_cb* fcb = mid->finalize_cb;
+    while(fcb) {
+        (fcb->callback)(fcb->uargs);
+        struct margo_finalize_cb* tmp = fcb;
+        fcb = fcb->next;
+        free(tmp);
+    }
+
     margo_timer_instance_finalize(mid);
 
     /* delete the hash used for multiplexing */
@@ -399,6 +417,23 @@ void margo_wait_for_finalize(margo_instance_id mid)
     return;
 }
 
+void margo_push_finalize_callback(
+            margo_instance_id mid,
+            void(*cb)(void*),                  
+            void* uargs)
+{
+    if(cb == NULL) return;
+
+    struct margo_finalize_cb* fcb = 
+        (struct margo_finalize_cb*)malloc(sizeof(*fcb));
+    fcb->callback = cb;
+    fcb->uargs = uargs;
+
+    struct margo_finalize_cb* next = mid->finalize_cb;
+    fcb->next = next;
+    mid->finalize_cb = fcb;
+}
+
 hg_id_t margo_register_name(margo_instance_id mid, const char *func_name,
     hg_proc_cb_t in_proc_cb, hg_proc_cb_t out_proc_cb, hg_rpc_cb_t rpc_cb)
 {