diff --git a/tools/import_sanef_databases.py b/tools/import_sanef_databases.py
new file mode 100644
index 0000000..ebbb1ac
--- /dev/null
+++ b/tools/import_sanef_databases.py
@@ -0,0 +1,144 @@
+"""Import iTop 'Instance de base de donnees' -> table server_databases.
+
+Cree la table si absente. Pour chaque ligne CSV, match sur Serveur->Nom
+(hostname) puis insert un enregistrement instance BDD avec:
+ - instance_name (Nom)
+ - db_type (Logiciel->Nom: SAP Hana, Oracle, PostgreSQL, SQL Server, MySQL...)
+ - db_version (Database_version / Logiciel->Version)
+ - db_edition (Database édition)
+ - cluster_name (Cluster id->Nom)
+ - environnement / etat iTop
+ - description
+
+Usage:
+    python tools/import_sanef_databases.py <csv> [--truncate] [--dry-run]
+"""
+import os
+import csv
+import argparse
+from sqlalchemy import create_engine, text
+
+DATABASE_URL = os.getenv("DATABASE_URL_DEMO") or os.getenv("DATABASE_URL") \
+    or "postgresql://patchcenter:PatchCenter2026!@localhost:5432/patchcenter_demo"
+
+
+def clean(v):
+    if v is None:
+        return None
+    s = str(v).strip()
+    return s or None
+
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("csv_path")
+    parser.add_argument("--truncate", action="store_true",
+                        help="Vide server_databases avant import")
+    parser.add_argument("--dry-run", action="store_true")
+    args = parser.parse_args()
+
+    engine = create_engine(DATABASE_URL)
+    print(f"[INFO] DB: {DATABASE_URL.split('@')[-1]}")
+    print(f"[INFO] CSV: {args.csv_path}")
+    conn = engine.connect().execution_options(isolation_level="AUTOCOMMIT")
+
+    # Cree la table si absente
+    conn.execute(text("""
+        CREATE TABLE IF NOT EXISTS server_databases (
+            id serial PRIMARY KEY,
+            server_id integer REFERENCES servers(id) ON DELETE CASCADE,
+            hostname citext,
+            instance_name varchar(100),
+            db_type varchar(50),
+            db_version varchar(100),
+            db_edition varchar(50),
+            cluster_name varchar(100),
+            environnement varchar(50),
+            etat varchar(30),
+            description text,
+            created_at timestamptz DEFAULT now(),
+            UNIQUE (hostname, instance_name)
+        )
+    """))
+    conn.execute(text("CREATE INDEX IF NOT EXISTS idx_srv_db_server ON server_databases(server_id)"))
+    conn.execute(text("CREATE INDEX IF NOT EXISTS idx_srv_db_type ON server_databases(db_type)"))
+
+    if args.truncate and not args.dry_run:
+        conn.execute(text("TRUNCATE server_databases RESTART IDENTITY CASCADE"))
+        print("[INFO] Table vidée")
+
+    with open(args.csv_path, "r", encoding="utf-8-sig", newline="") as f:
+        sample = f.read(4096); f.seek(0)
+        delim = ";" if sample.count(";") > sample.count(",") else ","
+        rows = list(csv.DictReader(f, delimiter=delim))
+    print(f"[INFO] {len(rows)} lignes (delim={delim!r})")
+
+    stats = {"inserted": 0, "updated": 0, "no_server": 0, "skipped": 0}
+
+    for r in rows:
+        hostname = clean(r.get("Serveur->Nom") or "")
+        if hostname:
+            hostname = hostname.split(".")[0].lower()
+        instance = clean(r.get("Nom"))
+        if not hostname or not instance:
+            stats["skipped"] += 1
+            continue
+
+        db_type = clean(r.get("Logiciel->Nom") or r.get("Type"))
+        db_version = clean(r.get("Database_version") or r.get("Logiciel->Version")
+                           or r.get("Version complète avec niveau de patch"))
+        db_edition = clean(r.get("Database édition"))
+        cluster = clean(r.get("Cluster id->Nom"))
+        env = clean(r.get("Environnement"))
+        etat = clean(r.get("Etat"))
+        desc = clean(r.get("Description"))
+
+        if db_version:
+            db_version = db_version[:100]
+        if db_type:
+            db_type = db_type[:50]
+
+        # Retrouve server_id
+        srv = conn.execute(text("SELECT id FROM servers WHERE hostname=:h"),
+                           {"h": hostname}).fetchone()
+        sid = srv.id if srv else None
+        if not sid:
+            stats["no_server"] += 1
+
+        if args.dry_run:
+            print(f"  DRY: {hostname:20s} {instance:20s} [{db_type or '?'}]")
+            continue
+
+        # Upsert via ON CONFLICT (hostname, instance_name)
+        try:
+            conn.execute(text("""
+                INSERT INTO server_databases
+                (server_id, hostname, instance_name, db_type, db_version,
+                 db_edition, cluster_name, environnement, etat, description)
+                VALUES (:sid, :h, :inst, :type, :ver, :ed, :cl, :env, :etat, :desc)
+                ON CONFLICT (hostname, instance_name) DO UPDATE SET
+                    server_id = EXCLUDED.server_id,
+                    db_type = EXCLUDED.db_type,
+                    db_version = EXCLUDED.db_version,
+                    db_edition = EXCLUDED.db_edition,
+                    cluster_name = EXCLUDED.cluster_name,
+                    environnement = EXCLUDED.environnement,
+                    etat = EXCLUDED.etat,
+                    description = EXCLUDED.description
+            """), {
+                "sid": sid, "h": hostname, "inst": instance,
+                "type": db_type, "ver": db_version, "ed": db_edition,
+                "cl": cluster, "env": env, "etat": etat, "desc": desc,
+            })
+            stats["inserted"] += 1
+        except Exception as e:
+            print(f"  [ERR] {hostname}/{instance}: {str(e)[:150]}")
+            stats["skipped"] += 1
+
+    conn.close()
+    print(f"\n[DONE] Inserts/upserts: {stats['inserted']} | "
+          f"Sans serveur en base: {stats['no_server']} | Skip: {stats['skipped']}")
+
+
+if __name__ == "__main__":
+    main()