diff --git a/integrations/adk-middleware/python/examples/server/api/custom_header_route.py b/integrations/adk-middleware/python/examples/server/api/custom_header_route.py
new file mode 100644
index 000000000..723527387
--- /dev/null
+++ b/integrations/adk-middleware/python/examples/server/api/custom_header_route.py
@@ -0,0 +1,91 @@
+"""Custom FastAPI route that pulls a user id from a request header.
+
+This example shows how to keep the default ADK route while adding a
+custom route that injects a header value (``x-user-id``) into the
+``RunAgentInput.state`` before forwarding the request to the ADK agent.
+The ``user_id_extractor`` reads the injected value so session handling can
+still use the custom user id.
+"""
+
+from __future__ import annotations
+
+from fastapi import FastAPI, Request
+from fastapi.responses import StreamingResponse
+
+from ag_ui_adk import ADKAgent, add_adk_fastapi_endpoint
+from ag_ui.core import EventType, RunAgentInput, RunErrorEvent
+from ag_ui.encoder import EventEncoder
+from google.adk.agents import LlmAgent
+
+
+def user_id_extractor(input_data: RunAgentInput) -> str:
+    """Pull the user id from state when present."""
+    if input_data.state and "user_id" in input_data.state:
+        return str(input_data.state["user_id"])
+    return "anonymous"
+
+
+# Create a sample ADK agent (this would be your actual agent)
+sample_agent = LlmAgent(
+    name="assistant",
+    model="gemini-2.0-flash",
+    instruction="""
+    You are a helpful assistant. You should respond succinctly while
+    honoring the requested user id when it is provided.
+    """,
+)
+
+# Create ADK middleware agent instance
+chat_agent = ADKAgent(
+    adk_agent=sample_agent,
+    app_name="demo_app",
+    user_id_extractor=user_id_extractor,
+    session_timeout_seconds=3600,
+    use_in_memory_services=True,
+)
+
+# Create FastAPI app
+app = FastAPI(title="ADK Middleware Header-based User Id")
+
+# Keep the default ADK endpoint for reference
+add_adk_fastapi_endpoint(app, chat_agent, path="/")
+
+
+async def stream_agent_response(input_data: RunAgentInput, request: Request) -> StreamingResponse:
+    """Mirror the built-in ADK endpoint's streaming behavior.
+
+    A generator is still required because the ADK agent yields a stream of
+    protocol events; this helper just keeps the streaming logic in one place so
+    both routes can share it.
+    """
+
+    encoder = EventEncoder(accept=request.headers.get("accept"))
+
+    async def event_generator():
+        try:
+            async for event in chat_agent.run(input_data):
+                yield encoder.encode(event)
+        except Exception as agent_error:
+            error_event = RunErrorEvent(
+                type=EventType.RUN_ERROR,
+                message=f"Agent execution failed: {agent_error}",
+                code="AGENT_ERROR",
+            )
+            yield encoder.encode(error_event)
+
+    return StreamingResponse(event_generator(), media_type=encoder.get_content_type())
+
+
+@app.post("/with-header")
+async def chat_with_header(input_data: RunAgentInput, request: Request):
+    """ADK route that adds ``x-user-id`` into state before running the agent."""
+
+    # Copy the incoming state to avoid mutating pydantic internals in place
+    input_state = dict(input_data.state or {})
+
+    header_user_id = request.headers.get("x-user-id")
+    if header_user_id:
+        input_state["user_id"] = header_user_id
+        input_data.state = input_state
+
+    return await stream_agent_response(input_data, request)