diff --git a/connectorx-python/connectorx/__init__.py b/connectorx-python/connectorx/__init__.py
index 8415c5126..1f91fdc9d 100644
--- a/connectorx-python/connectorx/__init__.py
+++ b/connectorx-python/connectorx/__init__.py
@@ -1,6 +1,6 @@
 from __future__ import annotations
 
-from typing import Any
+from typing import Any, Literal, TYPE_CHECKING, overload
 
 from importlib.metadata import version
 
@@ -11,6 +11,13 @@
     get_meta as _get_meta,
 )
 
+if TYPE_CHECKING:
+    import pandas as pd
+    import polars as pl
+    import modin.pandas as mpd
+    import dask.dataframe as dd
+    import pyarrow as pa
+
 __version__ = version(__name__)
 
 import os
@@ -27,8 +34,10 @@
     "CX_REWRITER_PATH", os.path.join(dir_path, "dependencies/federated-rewriter.jar")
 )
 
+Protocol = Literal["csv", "binary", "cursor", "simple", "text"]
 
-def rewrite_conn(conn: str, protocol: str | None = None):
+
+def rewrite_conn(conn: str, protocol: Protocol | None = None) -> tuple[str, Protocol]:
     if not protocol:
         # note: redshift/clickhouse are not compatible with the 'binary' protocol, and use other database
         # drivers to connect. set a compatible protocol and masquerade as the appropriate backend.
@@ -47,8 +56,8 @@ def rewrite_conn(conn: str, protocol: str | None = None):
 def get_meta(
     conn: str,
     query: str,
-    protocol: str | None = None,
-):
+    protocol: Protocol | None = None,
+) -> pd.DataFrame:
     """
     Get metadata (header) of the given query (only for pandas)
 
@@ -75,7 +84,7 @@ def partition_sql(
     partition_on: str,
     partition_num: int,
     partition_range: tuple[int, int] | None = None,
-):
+) -> list[str]:
     """
     Partition the sql query
 
@@ -106,11 +115,11 @@ def read_sql_pandas(
     sql: list[str] | str,
     con: str | dict[str, str],
     index_col: str | None = None,
-    protocol: str | None = None,
+    protocol: Protocol | None = None,
     partition_on: str | None = None,
     partition_range: tuple[int, int] | None = None,
     partition_num: int | None = None,
-):
+) -> pd.DataFrame:
     """
     Run the SQL query, download the data from database into a dataframe.
     First several parameters are in the same name and order with `pandas.read_sql`.
@@ -142,17 +151,103 @@ def read_sql_pandas(
     )
 
 
+# default return pd.DataFrame
+@overload
 def read_sql(
     conn: str | dict[str, str],
     query: list[str] | str,
     *,
-    return_type: str = "pandas",
-    protocol: str | None = None,
+    protocol: Protocol | None = None,
     partition_on: str | None = None,
     partition_range: tuple[int, int] | None = None,
     partition_num: int | None = None,
     index_col: str | None = None,
-):
+) -> pd.DataFrame: ...
+
+
+@overload
+def read_sql(
+    conn: str | dict[str, str],
+    query: list[str] | str,
+    *,
+    return_type: Literal["pandas"],
+    protocol: Protocol | None = None,
+    partition_on: str | None = None,
+    partition_range: tuple[int, int] | None = None,
+    partition_num: int | None = None,
+    index_col: str | None = None,
+) -> pd.DataFrame: ...
+
+
+@overload
+def read_sql(
+    conn: str | dict[str, str],
+    query: list[str] | str,
+    *,
+    return_type: Literal["arrow", "arrow2"],
+    protocol: Protocol | None = None,
+    partition_on: str | None = None,
+    partition_range: tuple[int, int] | None = None,
+    partition_num: int | None = None,
+    index_col: str | None = None,
+) -> pa.Table: ...
+
+
+@overload
+def read_sql(
+    conn: str | dict[str, str],
+    query: list[str] | str,
+    *,
+    return_type: Literal["modin"],
+    protocol: Protocol | None = None,
+    partition_on: str | None = None,
+    partition_range: tuple[int, int] | None = None,
+    partition_num: int | None = None,
+    index_col: str | None = None,
+) -> mpd.DataFrame: ...
+
+
+@overload
+def read_sql(
+    conn: str | dict[str, str],
+    query: list[str] | str,
+    *,
+    return_type: Literal["dask"],
+    protocol: Protocol | None = None,
+    partition_on: str | None = None,
+    partition_range: tuple[int, int] | None = None,
+    partition_num: int | None = None,
+    index_col: str | None = None,
+) -> dd.DataFrame: ...
+
+
+@overload
+def read_sql(
+    conn: str | dict[str, str],
+    query: list[str] | str,
+    *,
+    return_type: Literal["polars", "polars2"],
+    protocol: Protocol | None = None,
+    partition_on: str | None = None,
+    partition_range: tuple[int, int] | None = None,
+    partition_num: int | None = None,
+    index_col: str | None = None,
+) -> pl.DataFrame: ...
+
+
+def read_sql(
+    conn: str | dict[str, str],
+    query: list[str] | str,
+    *,
+    return_type: Literal[
+        "pandas", "polars", "polars2", "arrow", "arrow2", "modin", "dask"
+    ] = "pandas",
+    protocol: Protocol | None = None,
+    partition_on: str | None = None,
+    partition_range: tuple[int, int] | None = None,
+    partition_num: int | None = None,
+    index_col: str | None = None,
+) -> pd.DataFrame | mpd.DataFrame | dd.DataFrame | pl.DataFrame | pa.Table:
     """
     Run the SQL query, download the data from database into a dataframe.
 
@@ -318,7 +413,9 @@ def read_sql(
     return df
 
 
-def reconstruct_arrow(result: tuple[list[str], list[list[tuple[int, int]]]]):
+def reconstruct_arrow(
+    result: tuple[list[str], list[list[tuple[int, int]]]],
+) -> pa.Table:
     import pyarrow as pa
 
     names, ptrs = result
@@ -334,7 +431,7 @@ def reconstruct_arrow(result: tuple[list[str], list[list[tuple[int, int]]]]):
     return pa.Table.from_batches(rbs)
 
 
-def reconstruct_pandas(df_infos: dict[str, Any]):
+def reconstruct_pandas(df_infos: dict[str, Any]) -> pd.DataFrame:
     import pandas as pd
 
     data = df_infos["data"]
@@ -388,6 +485,6 @@ def remove_ending_semicolon(query: str) -> str:
       SQL query
 
     """
-    if query.endswith(';'):
+    if query.endswith(";"):
         query = query[:-1]
     return query