Merge pull request #209 from biggraph/darabos-mcp-fixes

examples/uploads/lynx_chatbot_scenario_selector.yaml

@@ -296,7 +296,4 @@
   link_answer: *link
   min_similarity_score: -1
 - name: malicious
-  mode: fixed_answer  # Could be sticky, but if we want the user to rephrase, let's give 2nd chance
-  fixed_answer: >
-    I am sorry, but I feel you want me use in a wrong way. If I feel it wrong, please try to
-    rephrase your question.
+  mode: llm_only

lynxkite-app/web/src/index.css

@@ -325,6 +325,7 @@ body {
   .table-viewer {
     td {
       padding: 5px 10px;
+      vertical-align: top;
     }
 
     .image-in-table {

lynxkite-bio/pyproject.toml

@@ -7,6 +7,7 @@ requires-python = ">=3.11"
 dependencies = [
     "fsspec>=2025.2.0",
     "joblib>=1.4.2",
+    "kubernetes>=32.0.1",
     "lynxkite-core",
     "lynxkite-graph-analytics",
     "pandas>=2.2.3",

lynxkite-bio/src/lynxkite_bio/k8s.py

@@ -19,7 +19,7 @@ Use `k8s.needs()` to declare a Kubernetes dependency for an operation. For examp
     port=8000,
     args=["--model", "google/gemma-3-1b-it"],
     health_probe="/health",
-    env=k8s.env_vars("HUGGING_FACE_HUB_TOKEN"),
+    forward_env=["HUGGING_FACE_HUB_TOKEN"],
     storage_path="/root/.cache/huggingface",
     storage_size="10Gi",
 )
@@ -39,8 +39,6 @@ import httpx
 from kubernetes import client, config
 from kubernetes.client.rest import ApiException
 
-config.load_kube_config()
-
 
 def _run(
     *,
@@ -51,6 +49,7 @@ def _run(
     storage_size,
     storage_path,
     health_probe,
+    forward_env,
     **kwargs,
 ):
     print(f"Starting {name} in namespace {namespace}...")
@@ -74,6 +73,10 @@ def _run(
                 ),
             )
         )
+    # Forward local environment variables to the container.
+    kwargs.setdefault("env", []).extend(
+        [{"name": name, "value": os.environ[name]} for name in forward_env]
+    )
     container = client.V1Container(
         name=name,
         image=image,
@@ -194,17 +197,13 @@ def _pvc_exists(name: str, namespace: str = "default") -> bool:
             raise
 
 
-def env_vars(*names: str):
-    """A convenient way to pass local environment variables to the microservice."""
-    return [{"name": name, "value": os.environ[name]} for name in names]
-
-
 def needs(
     name: str,
     image: str,
     port: int,
     args: list = None,
     env: list = None,
+    forward_env: list = None,
     health_probe: str = None,
     storage_size: str = None,
     storage_path: str = "/data",
@@ -217,12 +216,14 @@ def needs(
     def decorator(func):
         @functools.wraps(func)
         def wrapper(*func_args, **func_kwargs):
+            config.load_kube_config()
             _using(
                 name=name,
                 image=image,
                 port=port,
                 args=args or [],
                 env=env or [],
+                forward_env=forward_env or [],
                 health_probe=health_probe,
                 storage_size=storage_size,
                 storage_path=storage_path,

lynxkite-bio/src/lynxkite_bio/llm.py

@@ -22,7 +22,7 @@ op = ops.op_registration(ENV)
     port=8000,
     args=["--model", "google/gemma-3-1b-it"],
     health_probe="/health",
-    env=k8s.env_vars("HUGGING_FACE_HUB_TOKEN"),
+    forward_env=["HUGGING_FACE_HUB_TOKEN"],
     storage_path="/root/.cache/huggingface",
     storage_size="10Gi",
 )

lynxkite-bio/src/lynxkite_bio/nims.py

@@ -3,7 +3,9 @@
 from lynxkite_graph_analytics import Bundle
 from lynxkite.core import ops
 import httpx
+import io
 import pandas as pd
+import rdkit
 import os
 
 from . import k8s
@@ -122,14 +124,15 @@ def view_molecule(
     row_index: int = 0,
 ):
     molecule_data = bundle.dfs[molecule_table][molecule_column].iloc[row_index]
+    if isinstance(molecule_data, rdkit.Chem.Mol):
+        sio = io.StringIO()
+        with rdkit.Chem.SDWriter(sio) as w:
+            w.write(molecule_data)
+        molecule_data = sio.getvalue()
 
     return {
         "data": molecule_data,
-        "format": "pdb"
-        if molecule_data.startswith("ATOM")
-        else "sdf"
-        if molecule_data.startswith("CTfile")
-        else "smiles",
+        "format": "pdb" if molecule_data.startswith("ATOM") else "sdf",
     }
 
 

lynxkite-core/src/lynxkite/core/ops.py

@@ -254,8 +254,8 @@ def op(
             _view = "image"
             func = matplotlib_to_image(func)
         if slow:
-            func = mem.cache(func)
             func = make_async(func)
+            func = mem.cache(func)
         # Positional arguments are inputs.
         inputs = [
             Input(name=name, type=param.annotation)
@@ -456,6 +456,7 @@ def parse_doc(func):
         return doc
     if doc is None:
         return None
+    griffe.logger.setLevel("ERROR")
     ds = griffe.Docstring(doc, parent=_get_griffe_function(func))
     if "----" in doc:
         ds = ds.parse("numpy")

lynxkite-core/src/lynxkite/core/workspace.py

@@ -52,7 +52,7 @@ class WorkspaceNode(BaseConfig):
         """Notifies the frontend that work has started on this node."""
         self.data.error = None
         self.data.status = NodeStatus.active
-        if hasattr(self, "_crdt"):
+        if hasattr(self, "_crdt") and "data" in self._crdt:
             with self._crdt.doc.transaction():
                 self._crdt["data"]["error"] = None
                 self._crdt["data"]["status"] = NodeStatus.active
@@ -63,7 +63,7 @@ class WorkspaceNode(BaseConfig):
         self.data.input_metadata = result.input_metadata
         self.data.error = result.error
         self.data.status = NodeStatus.done
-        if hasattr(self, "_crdt"):
+        if hasattr(self, "_crdt") and "data" in self._crdt:
             with self._crdt.doc.transaction():
                 try:
                     self._crdt["data"]["status"] = NodeStatus.done

lynxkite-graph-analytics/src/lynxkite_graph_analytics/core.py

@@ -227,13 +227,13 @@ async def _execute_node(
             if p.type == nx.Graph:
                 if isinstance(x, Bundle):
                     x = x.to_nx()
-                assert isinstance(x, nx.Graph), "Input must be a graph."
+                assert isinstance(x, nx.Graph), f"Input must be a graph. Got: {x}"
             elif p.type == Bundle:
                 if isinstance(x, nx.Graph):
                     x = Bundle.from_nx(x)
                 elif isinstance(x, pd.DataFrame):
                     x = Bundle.from_df(x)
-                assert isinstance(x, Bundle), "Input must be a graph or dataframe."
+                assert isinstance(x, Bundle), f"Input must be a graph or dataframe. Got: {x}"
             inputs.append(x)
     except Exception as e:
         if not os.environ.get("LYNXKITE_SUPPRESS_OP_ERRORS"):

lynxkite-graph-analytics/src/lynxkite_graph_analytics/networkx_ops.py

@@ -156,7 +156,7 @@ def wrapped(name: str, func):
         for k, v in kwargs.items():
             if v == "None":
                 kwargs[k] = None
-        res = await ops.run_in_thread(func)(*args, **kwargs)
+        res = await ops.make_async(func)(*args, **kwargs)
         # Figure out what the returned value is.
         if isinstance(res, nx.Graph):
             return res

lynxkite-lynxscribe/src/lynxkite_lynxscribe/lynxscribe_ops.py

@@ -864,11 +864,15 @@ async def get_chat_api(ws: str):
     from lynxkite.core import workspace
 
     cwd = pathlib.Path()
-    path = cwd / ws
+    path = cwd / (ws + ".lynxkite.json")
     assert path.is_relative_to(cwd), f"Path '{path}' is invalid"
     assert path.exists(), f"Workspace {path} does not exist"
     ws = workspace.Workspace.load(path)
-    contexts = await ops.EXECUTORS[ENV](ws)
+    # Remove any test nodes.
+    ws.nodes = [op for op in ws.nodes if op.data.title != "Test Chat API"]
+    ws.normalize()
+    executor = ops.EXECUTORS[ENV]
+    contexts = await executor(ws)
     nodes = [op for op in ws.nodes if op.data.title == "LynxScribe RAG Graph Chatbot Backend"]
     [node] = nodes
     context = contexts[node.id]
@@ -879,9 +883,18 @@ async def stream_chat_api_response(request):
     chat_api = await get_chat_api(request["model"])
     request = ChatCompletionPrompt(**request)
     async for chunk in await chat_api.answer(request, stream=True):
+        chunk.sources = []
         yield chunk.model_dump_json()
 
 
+async def get_chat_api_response(request):
+    chat_api = await get_chat_api(request["model"])
+    request = ChatCompletionPrompt(**request)
+    response = await chat_api.answer(request, stream=False)
+    response.sources = []
+    return response.model_dump_json()
+
+
 async def api_service_post(request):
     """
     Serves a chat endpoint that matches LynxScribe's interface.
@@ -899,9 +912,12 @@ async def api_service_post(request):
     path = "/".join(request.url.path.split("/")[4:])
     request = await request.json()
     if path == "chat/completions":
-        from sse_starlette.sse import EventSourceResponse
+        if request["stream"]:
+            from sse_starlette.sse import EventSourceResponse
 
-        return EventSourceResponse(stream_chat_api_response(request))
+            return EventSourceResponse(stream_chat_api_response(request))
+        else:
+            return await get_chat_api_response(request)
     return {"error": "Not found"}
 
 
@@ -912,7 +928,7 @@ async def api_service_get(request):
             "object": "list",
             "data": [
                 {
-                    "id": ws,
+                    "id": ws.removesuffix(".lynxkite.json"),
                     "object": "model",
                     "created": 0,
                     "owned_by": "lynxkite",
@@ -924,7 +940,7 @@ async def api_service_get(request):
     return {"error": "Not found"}
 
 
-def get_lynxscribe_workspaces():
+def get_lynxscribe_workspaces() -> list[str]:
     from lynxkite.core import workspace
 
     workspaces = []
@@ -933,7 +949,7 @@ def get_lynxscribe_workspaces():
             try:
                 ws = workspace.Workspace.load(p)
                 if ws.env == ENV:
-                    workspaces.append(p)
+                    workspaces.append(str(p))
             except Exception:
                 pass  # Ignore files that are not valid workspaces.
     workspaces.sort()