Skip to content

RAGFlow is an agent-native database: a Retrieval-Augmented Generation (RAG) engine that grew into context infrastructure.

fix(mcp): include dataset name in list_datasets output (#17133) (#18710)

Harsh Kashyap7a549a073f8e
diff --git a/mcp/server/server.py b/mcp/server/server.py
index 07c74e28f8..d9453f58f1 100644
--- a/mcp/server/server.py
+++ b/mcp/server/server.py
@@ -242,7 +242,7 @@ class RAGFlowConnector:
result_list = []
for data in datasets:
- d = {"description": data["description"], "id": data["id"]}
+ d = {"id": data["id"], "name": data.get("name"), "description": data.get("description")}
result_list.append(json.dumps(d, ensure_ascii=False))
return "\n".join(result_list)
diff --git a/test/unit_test/mcp/test_server_datasets.py b/test/unit_test/mcp/test_server_datasets.py
index bd75e157cb..b7e0bc6a60 100644
--- a/test/unit_test/mcp/test_server_datasets.py
+++ b/test/unit_test/mcp/test_server_datasets.py
@@ -39,8 +39,14 @@ class _FakeResponse:
return self._payload
-def _datasets(count):
- return [{"id": f"dataset-{idx}", "description": f"description-{idx}"} for idx in range(count)]
+def _datasets(count, *, with_name=True):
+ out = []
+ for idx in range(count):
+ row = {"id": f"dataset-{idx}", "description": f"description-{idx}"}
+ if with_name:
+ row["name"] = f"name-{idx}"
+ out.append(row)
+ return out
@pytest.fixture()
@@ -115,3 +121,56 @@ async def test_list_datasets_clamps_explicit_page_size_to_rest_limit_and_preserv
"id": "dataset-1",
"name": "target",
}
+
+
+async def test_list_datasets_includes_name_field(monkeypatch, mcp_server):
+ """``list_datasets`` must surface each dataset's ``name`` so MCP clients
+ can show a human-readable label next to the opaque ``id``.
+
+ Regression for #17133: pre-fix the per-row dict was
+ ``{"description": ..., "id": ...}`` and dropped the ``name`` that
+ the REST API returns. The sibling ``list_chats`` already
+ includes ``id`` / ``name`` / ``description`` — the datasets
+ branch was the inconsistency.
+ """
+ connector = mcp_server.RAGFlowConnector(base_url=mcp_server.BASE_URL)
+ _stub_dataset_pages(monkeypatch, connector, _datasets(3))
+
+ result = await connector.list_datasets(api_key="unit-key")
+
+ rows = [json.loads(line) for line in result.splitlines()]
+ assert len(rows) == 3
+ # Every row has the three fields the REST API returns.
+ for row in rows:
+ assert set(row.keys()) == {"id", "name", "description"}, f"row {row!r} is missing one of id/name/description"
+ # Names round-trip in document order.
+ assert [row["name"] for row in rows] == ["name-0", "name-1", "name-2"]
+ assert [row["id"] for row in rows] == ["dataset-0", "dataset-1", "dataset-2"]
+ assert [row["description"] for row in rows] == [
+ "description-0",
+ "description-1",
+ "description-2",
+ ]
+
+
+async def test_list_datasets_name_omitted_when_missing(monkeypatch, mcp_server):
+ """If the REST API does not include ``name`` for some row (older
+ dataset rows, partial records, migration backlog), the per-row
+ dict must still serialise with ``name`` set to ``None`` so the
+ shape is stable. The client can then show a fallback label.
+ """
+ connector = mcp_server.RAGFlowConnector(base_url=mcp_server.BASE_URL)
+ # No ``name`` field on any row — matches a real API response
+ # where the field is absent.
+ _stub_dataset_pages(monkeypatch, connector, _datasets(2, with_name=False))
+
+ result = await connector.list_datasets(api_key="unit-key")
+
+ rows = [json.loads(line) for line in result.splitlines()]
+ assert len(rows) == 2
+ for row in rows:
+ assert "name" in row
+ assert row["name"] is None
+ assert set(row.keys()) == {"id", "name", "description"}