diff --git a/src/tracksdata/graph/_sql_graph.py b/src/tracksdata/graph/_sql_graph.py
index f0b9ab86..b4a64793 100644
--- a/src/tracksdata/graph/_sql_graph.py
+++ b/src/tracksdata/graph/_sql_graph.py
@@ -1,4 +1,6 @@
 import binascii
+import uuid
+import weakref
 from collections.abc import Callable, Sequence
 from enum import Enum
 from typing import TYPE_CHECKING, Any, TypeVar
@@ -56,6 +58,85 @@ def _data_numpy_to_native(data: dict[str, Any]) -> None:
             data[k] = v.item()
 
 
+# Module-level (not methods) so they can be registered with ``weakref.finalize``
+# without holding a bound reference to the owning object, which would prevent
+# it from ever being collected.
+def _drop_scratch_tables(engine: sa.Engine, tables: list[sa.Table]) -> None:
+    """Drop scratch tables, swallowing errors (e.g. at interpreter shutdown)."""
+    for table in tables:
+        try:
+            table.drop(engine)
+        except Exception as exc:
+            LOG.debug("Failed to drop scratch table %s: %s", table.name, exc)
+
+
+class _SqlIdSet:
+    """A set of ids usable in SQL ``IN`` clauses without overflowing bind limits.
+
+    Small sets compile to inline ``col.in_([...])``; larger sets are materialized
+    into a per-instance scratch table and matched via ``col.in_(SELECT id FROM
+    scratch)``. The same ``_SqlIdSet`` may be reused against multiple columns —
+    each call to :meth:`in_clause` emits a fresh expression backed by the same
+    underlying ids.
+
+    ``occurrences`` is the maximum number of times the id set will be expanded
+    in a single compiled statement (e.g. filtering both ``source_id`` and
+    ``target_id`` of an edge table counts as 2). The scratch-table cutoff is
+    divided by it so that ``len(ids) * occurrences`` stays safely under the
+    backend's bound-variable limit.
+    """
+
+    def __init__(
+        self,
+        graph: "SQLGraph",
+        ids: Sequence[int],
+        *,
+        occurrences: int = 1,
+    ) -> None:
+        if hasattr(ids, "tolist"):
+            ids = ids.tolist()
+        self._ids: list[int] = list(ids)
+        self._graph = graph
+
+        limit = max(1, graph._sql_chunk_size() // max(1, occurrences))
+        if len(self._ids) > limit:
+            self._scratch: sa.Table | None = graph._create_id_scratch_table(self._ids)
+        else:
+            self._scratch = None
+
+    @property
+    def ids(self) -> list[int]:
+        return self._ids
+
+    @property
+    def uses_scratch_table(self) -> bool:
+        return self._scratch is not None
+
+    def in_clause(self, column: sa.ColumnElement) -> sa.ColumnElement[bool]:
+        if self._scratch is None:
+            return column.in_(self._ids)
+        return column.in_(sa.select(self._scratch.c.id))
+
+    def close(self) -> None:
+        if self._scratch is not None:
+            _drop_scratch_tables(self._graph._engine, [self._scratch])
+            self._scratch = None
+
+    def __enter__(self) -> "_SqlIdSet":
+        return self
+
+    def __exit__(self, *exc: object) -> None:
+        self.close()
+
+
+def _close_id_sets(id_sets: list[_SqlIdSet]) -> None:
+    for id_set in id_sets:
+        try:
+            id_set.close()
+        except Exception as exc:
+            LOG.debug("Failed to close _SqlIdSet: %s", exc)
+
+
 def _filter_query(
     query: sa.Select,
     table: type[DeclarativeBase],
@@ -99,6 +180,7 @@ def __init__(
         self._node_attr_comps, self._edge_attr_comps = split_attr_comps(attr_filters)
         self._include_targets = include_targets
         self._include_sources = include_sources
+        self._id_sets: list[_SqlIdSet] = []
 
         # creating initial query
         self._node_query: sa.Select = sa.select(self._graph.Node)
@@ -106,18 +188,20 @@ def __init__(
         node_filtered = False
 
         if node_ids is not None:
-            if hasattr(node_ids, "tolist"):
-                node_ids = node_ids.tolist()
-
-            self._node_query = self._node_query.filter(self._graph.Node.node_id.in_(node_ids))
+            # The id set is expanded once for Node.node_id, plus once each for
+            # Edge.target_id / Edge.source_id when those filters are not
+            # suppressed by ``include_targets`` / ``include_sources``. The
+            # scratch-table cutoff is divided accordingly so that the total
+            # number of bound variables stays under the backend's limit.
+            occurrences = 1 + (not self._include_targets) + (not self._include_sources)
+            id_set = _SqlIdSet(self._graph, node_ids, occurrences=occurrences)
+            self._id_sets.append(id_set)
+
+            self._node_query = self._node_query.filter(id_set.in_clause(self._graph.Node.node_id))
             if not self._include_targets:
-                self._edge_query = self._edge_query.filter(
-                    self._graph.Edge.target_id.in_(node_ids),
-                )
+                self._edge_query = self._edge_query.filter(id_set.in_clause(self._graph.Edge.target_id))
             if not self._include_sources:
-                self._edge_query = self._edge_query.filter(
-                    self._graph.Edge.source_id.in_(node_ids),
-                )
+                self._edge_query = self._edge_query.filter(id_set.in_clause(self._graph.Edge.source_id))
             node_filtered = True
 
         if self._node_attr_comps:
@@ -182,6 +266,13 @@ def __init__(
 
             self._node_query = sa.union(*nodes_query)
 
+        if self._uses_scratch_tables():
+            weakref.finalize(self, _close_id_sets, self._id_sets)
+
+    def _uses_scratch_tables(self) -> bool:
+        """Whether any id set backing this filter materialized a scratch table."""
+        return any(id_set.uses_scratch_table for id_set in self._id_sets)
+
     @cache_method
     def node_ids(self) -> list[int]:
         """
@@ -1092,19 +1183,18 @@ def overlaps(
         """
         Get the overlaps between the nodes in `node_ids`.
         """
-        if hasattr(node_ids, "tolist"):
-            node_ids = node_ids.tolist()
-
         with Session(self._engine) as session:
             query = session.query(self.Overlap.source_id, self.Overlap.target_id)
 
-            if node_ids is not None:
+            if node_ids is None:
+                return [[source_id, target_id] for source_id, target_id in query.all()]
+
+            with _SqlIdSet(self, node_ids, occurrences=2) as id_set:
                 query = query.filter(
-                    self.Overlap.source_id.in_(node_ids),
-                    self.Overlap.target_id.in_(node_ids),
+                    id_set.in_clause(self.Overlap.source_id),
+                    id_set.in_clause(self.Overlap.target_id),
                 )
-
-            return [[source_id, target_id] for source_id, target_id in query.all()]
+                return [[source_id, target_id] for source_id, target_id in query.all()]
 
     def has_overlaps(self) -> bool:
         """
@@ -1794,6 +1884,33 @@ def _sql_chunk_size(self) -> int:
 
         return chunk_size
 
+    def _create_id_scratch_table(self, ids: Sequence[int]) -> sa.Table:
+        """Create a uniquely-named helper table holding ``ids``.
+
+        Used to work around SQL bound-variable limits when filtering by large
+        ``IN (...)`` lists: callers replace ``col.in_(ids)`` with
+        ``col.in_(sa.select(table.c.id))``. The caller owns the returned table
+        and is responsible for dropping it.
+        """
+        unique_ids = list({int(v) for v in ids})
+
+        name = f"_tracksdata_ids_{uuid.uuid4().hex}"
+        table = sa.Table(
+            name,
+            sa.MetaData(),
+            sa.Column("id", sa.BigInteger, primary_key=True),
+        )
+        table.create(self._engine)
+
+        chunk_size = max(1, self._sql_chunk_size())
+        with self._engine.begin() as conn:
+            for i in range(0, len(unique_ids), chunk_size):
+                conn.execute(
+                    table.insert(),
+                    [{"id": v} for v in unique_ids[i : i + chunk_size]],
+                )
+        return table
+
     def _update_table(
         self,
         table_class: type[DeclarativeBase],
@@ -2009,18 +2126,18 @@ def _get_degree(
                 return int(session.execute(stmt).scalar())
 
         stmt = sa.select(edge_key_col, sa.func.count()).group_by(edge_key_col)
-        if node_ids is not None:
-            stmt = stmt.where(edge_key_col.in_(node_ids))
-
-        with Session(self._engine) as session:
-            # get the number of edges for each using group by and count
-            degree = dict(session.execute(stmt).all())
 
         if node_ids is None:
-            # this is necessary to make sure it's the same order as node_ids
+            with Session(self._engine) as session:
+                degree = dict(session.execute(stmt).all())
+            # preserve the canonical node ordering
             return [degree.get(node_id, 0) for node_id in self.node_ids()]
 
-        return [degree.get(node_id, 0) for node_id in node_ids]
+        with _SqlIdSet(self, node_ids, occurrences=1) as id_set:
+            stmt = stmt.where(id_set.in_clause(edge_key_col))
+            with Session(self._engine) as session:
+                degree = dict(session.execute(stmt).all())
+            return [degree.get(node_id, 0) for node_id in id_set.ids]
 
     def in_degree(self, node_ids: list[int] | int | None = None) -> list[int] | int:
         """
diff --git a/src/tracksdata/graph/_test/test_subgraph.py b/src/tracksdata/graph/_test/test_subgraph.py
index 96cbfa1a..89484803 100644
--- a/src/tracksdata/graph/_test/test_subgraph.py
+++ b/src/tracksdata/graph/_test/test_subgraph.py
@@ -1,3 +1,4 @@
+import itertools
 import re
 from collections.abc import Callable
 from contextlib import contextmanager
@@ -1302,3 +1303,89 @@ def test_edge_list(graph_backend: BaseGraph, use_subgraph: bool) -> None:
         )
     )
     assert edge_list == expected_edge_list
+
+
+def _build_chain_graph(graph: SQLGraph, n_nodes: int) -> list[int]:
+    node_ids: list[int] = []
+    for t in range(n_nodes):
+        node_ids.append(graph.add_node({DEFAULT_ATTR_KEYS.T: t}))
+    for src, tgt in itertools.pairwise(node_ids):
+        graph.add_edge(src, tgt, {})
+    graph.add_overlap(node_ids[0], node_ids[1])
+    graph.add_overlap(node_ids[2], node_ids[3])
+    return node_ids
+
+
+def _scratch_table_count(graph: SQLGraph) -> int:
+    """Count leftover ``_tracksdata_ids_*`` scratch tables in a SQLite graph."""
+    import sqlalchemy as sa
+
+    with graph._engine.connect() as conn:
+        return conn.execute(
+            sa.text("SELECT COUNT(*) FROM sqlite_master WHERE type='table' AND name LIKE '_tracksdata_ids_%'")
+        ).scalar()
+
+
+def test_sql_graph_filter_large_node_ids(tmp_path, monkeypatch: pytest.MonkeyPatch) -> None:
+    """Filtering with more ids than SQLite's variable limit must not raise.
+
+    Reproduces the ``OperationalError: too many SQL variables`` failure by
+    forcing the scratch-table code path via a tiny chunk size.
+    """
+    graph = SQLGraph("sqlite", str(tmp_path / "scratch.db"))
+    n_nodes = 40
+    node_ids = _build_chain_graph(graph, n_nodes)
+
+    # Force scratch-table path on every call site.
+    monkeypatch.setattr(SQLGraph, "_sql_chunk_size", lambda self: 4)
+
+    # Context-manager paths (overlaps, _get_degree) must drop their scratch
+    # tables once the block exits — the count should return to baseline after
+    # each call regardless of whether the scratch path fired inside.
+    assert _scratch_table_count(graph) == 0
+    in_deg = graph.in_degree(node_ids)
+    assert _scratch_table_count(graph) == 0
+    out_deg = graph.out_degree(node_ids)
+    assert _scratch_table_count(graph) == 0
+    overlaps = graph.overlaps(node_ids)
+    assert _scratch_table_count(graph) == 0
+
+    assert sum(in_deg) == n_nodes - 1
+    assert sum(out_deg) == n_nodes - 1
+    assert sorted(map(tuple, overlaps)) == sorted([(node_ids[0], node_ids[1]), (node_ids[2], node_ids[3])])
+
+    filtered = graph.filter(node_ids=node_ids)
+    # Confirm the scratch-table code path was taken rather than raw IN (...).
+    assert filtered._uses_scratch_tables()
+    subgraph = filtered.subgraph()
+    assert subgraph.num_nodes() == n_nodes
+    assert subgraph.num_edges() == n_nodes - 1
+
+
+def test_sql_graph_filter_borderline_node_ids(tmp_path, monkeypatch: pytest.MonkeyPatch) -> None:
+    """The scratch cutoff must account for how many times ids appear per statement.
+
+    With ``_sql_chunk_size() == 12`` and ``SQLFilter`` using ``occurrences=3``,
+    a list of 5 ids would compile to ~15 bound variables — above the limit —
+    even though ``len(node_ids) <= chunk_size``. The helper must still switch
+    to the scratch-table path in that band.
+    """
+    graph = SQLGraph("sqlite", str(tmp_path / "scratch.db"))
+    n_nodes = 5
+    node_ids = _build_chain_graph(graph, n_nodes)
+
+    monkeypatch.setattr(SQLGraph, "_sql_chunk_size", lambda self: 12)
+
+    filtered = graph.filter(node_ids=node_ids)
+    # 5 ids fits under chunk_size=12 inline, but with occurrences=3 the
+    # effective cutoff is 12 // 3 == 4, so scratch must kick in.
+    assert filtered._uses_scratch_tables()
+    subgraph = filtered.subgraph()
+    assert subgraph.num_nodes() == n_nodes
+    assert subgraph.num_edges() == n_nodes - 1
+
+    # overlaps() uses occurrences=2 → cutoff 6, so len==5 stays inline.
+    # Still assert it returns the right data regardless of path.
+    assert sorted(map(tuple, graph.overlaps(node_ids))) == sorted(
+        [(node_ids[0], node_ids[1]), (node_ids[2], node_ids[3])]
+    )