diff --git a/swh/indexer/metadata.py b/swh/indexer/metadata.py
--- a/swh/indexer/metadata.py
+++ b/swh/indexer/metadata.py
@@ -3,8 +3,6 @@
 # License: GNU General Public License version 3, or any later version
 # See top-level LICENSE file for more information
 
-from copy import deepcopy
-
 from swh.core.utils import grouper
 
 from swh.indexer.codemeta import merge_documents
@@ -12,7 +10,6 @@
 from swh.indexer.origin_head import OriginHeadIndexer
 from swh.indexer.metadata_dictionary import MAPPINGS
 from swh.indexer.metadata_detector import detect_metadata
-from swh.indexer.storage import INDEXER_CFG_KEY
 
 from swh.model import hashutil
 
@@ -160,15 +157,21 @@
         }
 
         try:
+            # Get the root directory of the revision
             root_dir = rev['directory']
             dir_ls = list(self.storage.directory_ls(root_dir, recursive=False))
+
             if [entry['type'] for entry in dir_ls] == ['dir']:
                 # If the root is just a single directory, recurse into it
                 # eg. PyPI packages, GNU tarballs
                 subdir = dir_ls[0]['target']
                 dir_ls = self.storage.directory_ls(subdir, recursive=False)
+
+            # List files, and find which ones are metadata files
             files = [entry for entry in dir_ls if entry['type'] == 'file']
             detected_files = detect_metadata(files)
+
+            # Translate the metadata from these files
             (mappings, metadata) = self.translate_revision_intrinsic_metadata(
                 detected_files,
                 log_suffix='revision=%s' % hashutil.hash_to_hex(rev['id']))
@@ -197,6 +200,27 @@
         self.idx_storage.revision_intrinsic_metadata_add(
             results, conflict_update=(policy_update == 'update-dups'))
 
+    def get_cached_content_metadata(self, sha1s):
+        """Fetches the 'content_metadata' table to find metadata for
+        content indexed by the exact same tool as this one."""
+        cached_metadata = self.idx_storage.content_metadata_get(sha1s)
+        sha1_to_metadata = {}
+        for c in cached_metadata:
+            if c['tool']['id'] == self.tool['id']:
+                # If it was indexed by a different tool (or a different
+                # version of this tool), use the cache.
+                assert c['id'] not in sha1_to_metadata, (
+                    'Content %s has more than one metadata entry for tool %s.'
+                    % (hashutil.hash_to_hex(c['id']), self.tool))
+
+                sha1_to_metadata[c['id']] = c['metadata']
+            else:
+                # TODO: if same tool but older version, let's delete
+                # the cache?
+                pass
+
+        return sha1_to_metadata
+
     def translate_revision_intrinsic_metadata(
             self, detected_files, log_suffix):
         """
@@ -214,47 +238,24 @@
         """
         used_mappings = [MAPPINGS[context].name for context in detected_files]
         metadata = []
-        tool = {
-                'name': 'swh-metadata-translator',
-                'version': '0.0.2',
-                'configuration': {
-                },
-            }
-        # TODO: iterate on each context, on each file
-        # -> get raw_contents
-        # -> translate each content
-        config = {
-            k: self.config[k]
-            for k in [INDEXER_CFG_KEY, 'objstorage', 'storage']
-        }
-        config['tools'] = [tool]
         for context in detected_files.keys():
-            cfg = deepcopy(config)
-            cfg['tools'][0]['configuration']['context'] = context
-            c_metadata_indexer = ContentMetadataIndexer(config=cfg)
             # sha1s that are in content_metadata table
-            sha1s_in_storage = []
-            metadata_generator = self.idx_storage.content_metadata_get(
+            cached_metadata_for_files = self.get_cached_content_metadata(
                 detected_files[context])
-            for c in metadata_generator:
-                # extracting metadata
-                sha1 = c['id']
-                sha1s_in_storage.append(sha1)
-                local_metadata = c['metadata']
-                # local metadata is aggregated
-                if local_metadata:
-                    metadata.append(local_metadata)
+            metadata.extend(cached_metadata_for_files.values())
 
             sha1s_filtered = [item for item in detected_files[context]
-                              if item not in sha1s_in_storage]
+                              if item not in cached_metadata_for_files]
 
+            # If all the metadata are not already cached, run the
+            # ContentMetadataIndexer.
             if sha1s_filtered:
+                c_metadata_indexer = ContentMetadataIndexer(config=self.config)
                 # content indexing
                 try:
                     c_metadata_indexer.run(sha1s_filtered,
                                            policy_update='ignore-dups',
                                            log_suffix=log_suffix)
-                    # on the fly possibility:
                     for result in c_metadata_indexer.results:
                         local_metadata = result['metadata']
                         metadata.append(local_metadata)