diff --git a/docker/pyproject.deps.toml b/docker/pyproject.deps.toml
index c0eca25..abb03e5 100644
--- a/docker/pyproject.deps.toml
+++ b/docker/pyproject.deps.toml
@@ -1,6 +1,6 @@
 [project]
 name = "mcp-plex"
-version = "0.26.70"
+version = "0.26.71"
 requires-python = ">=3.11,<3.13"
 dependencies = [
   "fastmcp>=2.11.2",
diff --git a/mcp_plex/loader/pipeline/orchestrator.py b/mcp_plex/loader/pipeline/orchestrator.py
index 45363e7..8a51753 100644
--- a/mcp_plex/loader/pipeline/orchestrator.py
+++ b/mcp_plex/loader/pipeline/orchestrator.py
@@ -1 +1,177 @@
-"""Placeholder module for the loader pipeline."""
+"""Coordinating logic tying the loader pipeline stages together."""
+
+from __future__ import annotations
+
+import asyncio
+import inspect
+import logging
+from dataclasses import dataclass
+from typing import Awaitable, Callable
+
+from .channels import IngestQueue, PersistenceQueue
+
+LOGGER = logging.getLogger("mcp_plex.loader.orchestrator")
+
+
+@dataclass(frozen=True, slots=True)
+class _StageSpec:
+    """Descriptor for a running pipeline stage."""
+
+    role: str
+    worker_id: int | None = None
+
+
+class _StageFailure(Exception):
+    """Wrapper exception capturing the originating stage failure."""
+
+    def __init__(self, spec: _StageSpec, error: BaseException) -> None:
+        super().__init__(str(error))
+        self.spec = spec
+        self.error = error
+
+
+class LoaderOrchestrator:
+    """Run the ingestion, enrichment, and persistence stages with supervision."""
+
+    def __init__(
+        self,
+        *,
+        ingestion_stage: object,
+        enrichment_stage: object,
+        persistence_stage: object,
+        ingest_queue: IngestQueue,
+        persistence_queue: PersistenceQueue,
+        persistence_worker_count: int = 1,
+        logger: logging.Logger | None = None,
+    ) -> None:
+        if persistence_worker_count <= 0:
+            raise ValueError("persistence_worker_count must be positive")
+
+        self._ingestion_stage = ingestion_stage
+        self._enrichment_stage = enrichment_stage
+        self._persistence_stage = persistence_stage
+        self._ingest_queue = ingest_queue
+        self._persistence_queue = persistence_queue
+        self._persistence_worker_count = int(persistence_worker_count)
+        self._logger = logger or LOGGER
+
+    async def run(self) -> None:
+        """Execute the configured pipeline stages concurrently."""
+
+        try:
+            async with asyncio.TaskGroup() as group:
+                group.create_task(
+                    self._run_stage(
+                        _StageSpec(role="ingestion"),
+                        getattr(self._ingestion_stage, "run"),
+                    )
+                )
+                group.create_task(
+                    self._run_stage(
+                        _StageSpec(role="enrichment"),
+                        getattr(self._enrichment_stage, "run"),
+                    )
+                )
+                persistence_runner = getattr(self._persistence_stage, "run")
+                for worker_id in range(self._persistence_worker_count):
+                    group.create_task(
+                        self._run_stage(
+                            _StageSpec(role="persistence", worker_id=worker_id),
+                            persistence_runner,
+                            worker_id,
+                        )
+                    )
+        except* _StageFailure as exc_group:
+            failures = list(exc_group.exceptions)
+            await self._handle_failures(failures)
+            # Re-raise the first underlying error after cleanup so callers see the
+            # original exception rather than the wrapper.
+            raise failures[0].error
+
+    async def _run_stage(
+        self,
+        spec: _StageSpec,
+        runner: Callable[..., Awaitable[object] | object],
+        *args: object,
+    ) -> None:
+        """Execute *runner* and wrap unexpected exceptions with stage metadata."""
+
+        try:
+            result = runner(*args)
+            if inspect.isawaitable(result):
+                await result
+        except asyncio.CancelledError:
+            raise
+        except BaseException as exc:
+            raise _StageFailure(spec, exc) from exc
+
+    async def _handle_failures(self, failures: list[_StageFailure]) -> None:
+        """Log stage-specific failures and drain queues during cancellation."""
+
+        if not failures:
+            return
+
+        roles = {failure.spec.role for failure in failures}
+        if "ingestion" in roles:
+            self._logger.warning(
+                "Ingestion stage failed; cancelling enrichment and persistence tasks."
+            )
+        else:
+            self._logger.warning(
+                "Downstream stage failed; cancelling ingestion and related tasks."
+            )
+
+        for failure in failures:
+            stage_name = self._describe_stage(failure.spec)
+            self._logger.error(
+                "%s failed: %s",
+                stage_name,
+                failure.error,
+                exc_info=failure.error,
+            )
+
+        drained_ingest = self._drain_queue(self._ingest_queue)
+        drained_persist = self._drain_queue(self._persistence_queue)
+        if drained_ingest:
+            self._logger.debug(
+                "Drained %d item(s) from the ingest queue during cancellation.",
+                drained_ingest,
+            )
+        if drained_persist:
+            self._logger.debug(
+                "Drained %d item(s) from the persistence queue during cancellation.",
+                drained_persist,
+            )
+
+        # Yield to the event loop so cancelled tasks can finish cleanup before the
+        # caller observes the exception.  This mirrors the behaviour expected by
+        # the stage-specific tests which verify cancellation side-effects.
+        await asyncio.sleep(0)
+
+    def _drain_queue(self, queue: asyncio.Queue[object]) -> int:
+        """Remove any queued items so cancellation does not leave stale work."""
+
+        drained = 0
+        while True:
+            try:
+                queue.get_nowait()
+            except asyncio.QueueEmpty:
+                break
+            else:
+                drained += 1
+                try:
+                    queue.task_done()
+                except ValueError:  # Queue.join() not in use; ignore bookkeeping.
+                    pass
+        return drained
+
+    def _describe_stage(self, spec: _StageSpec) -> str:
+        """Return a human-friendly name for *spec*."""
+
+        role = spec.role.capitalize()
+        if spec.worker_id is None:
+            return f"{role} stage"
+        return f"{role} stage (worker {spec.worker_id})"
+
+
+__all__ = ["LoaderOrchestrator"]
diff --git a/pyproject.toml b/pyproject.toml
index 41b4c70..8b9f01e 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "mcp-plex"
-version = "0.26.70"
+version = "0.26.71"
 
 description = "Plex-Oriented Model Context Protocol Server"
 requires-python = ">=3.11,<3.13"
diff --git a/tests/test_loader_orchestrator.py b/tests/test_loader_orchestrator.py
new file mode 100644
index 0000000..42e24a4
--- /dev/null
+++ b/tests/test_loader_orchestrator.py
@@ -0,0 +1,166 @@
+import asyncio
+import logging
+
+import pytest
+
+from mcp_plex.loader.pipeline.orchestrator import LoaderOrchestrator
+
+
+class FailingIngestionStage:
+    def __init__(self, queue: asyncio.Queue[object]) -> None:
+        self.queue = queue
+
+    async def run(self) -> None:
+        await self.queue.put("batch-1")
+        raise RuntimeError("ingestion boom")
+
+
+class BlockingEnrichmentStage:
+    def __init__(self) -> None:
+        self.cancelled = asyncio.Event()
+        self._blocker = asyncio.Event()
+
+    async def run(self) -> None:
+        try:
+            await self._blocker.wait()
+        except asyncio.CancelledError:
+            self.cancelled.set()
+            raise
+
+
+class BlockingPersistenceStage:
+    def __init__(self, queue: asyncio.Queue[object]) -> None:
+        self.queue = queue
+        self.cancelled = asyncio.Event()
+
+    async def run(self, worker_id: int) -> None:
+        try:
+            while True:
+                await self.queue.get()
+        except asyncio.CancelledError:
+            self.cancelled.set()
+            raise
+
+
+class BlockingIngestionStage:
+    def __init__(self, queue: asyncio.Queue[object]) -> None:
+        self.queue = queue
+        self.cancelled = asyncio.Event()
+        self._blocker = asyncio.Event()
+
+    async def run(self) -> None:
+        try:
+            await self.queue.put("batch-1")
+            await self.queue.put("batch-2")
+            await self._blocker.wait()
+        except asyncio.CancelledError:
+            self.cancelled.set()
+            raise
+
+
+class SingleBatchEnrichmentStage:
+    def __init__(
+        self,
+        ingest_queue: asyncio.Queue[object],
+        persistence_queue: asyncio.Queue[object],
+    ) -> None:
+        self.ingest_queue = ingest_queue
+        self.persistence_queue = persistence_queue
+        self.cancelled = asyncio.Event()
+        self._blocker = asyncio.Event()
+
+    async def run(self) -> None:
+        try:
+            payload = await self.ingest_queue.get()
+            self.ingest_queue.task_done()
+            await self.persistence_queue.put(payload)
+            await self._blocker.wait()
+        except asyncio.CancelledError:
+            self.cancelled.set()
+            raise
+
+
+class FailingPersistenceStage:
+    def __init__(self, queue: asyncio.Queue[object]) -> None:
+        self.queue = queue
+
+    async def run(self, worker_id: int) -> None:
+        payload = await self.queue.get()
+        self.queue.task_done()
+        raise RuntimeError(f"persistence boom {worker_id}: {payload}")
+
+
+def _build_orchestrator(
+    *,
+    ingestion_stage: object,
+    enrichment_stage: object,
+    persistence_stage: object,
+    ingest_queue: asyncio.Queue[object],
+    persistence_queue: asyncio.Queue[object],
+) -> LoaderOrchestrator:
+    return LoaderOrchestrator(
+        ingestion_stage=ingestion_stage,
+        enrichment_stage=enrichment_stage,
+        persistence_stage=persistence_stage,
+        ingest_queue=ingest_queue,
+        persistence_queue=persistence_queue,
+        persistence_worker_count=1,
+    )
+
+
+def test_ingestion_failure_cancels_downstream(caplog: pytest.LogCaptureFixture) -> None:
+    ingest_queue: asyncio.Queue[object] = asyncio.Queue()
+    persistence_queue: asyncio.Queue[object] = asyncio.Queue()
+    ingestion_stage = FailingIngestionStage(ingest_queue)
+    enrichment_stage = BlockingEnrichmentStage()
+    persistence_stage = BlockingPersistenceStage(persistence_queue)
+    orchestrator = _build_orchestrator(
+        ingestion_stage=ingestion_stage,
+        enrichment_stage=enrichment_stage,
+        persistence_stage=persistence_stage,
+        ingest_queue=ingest_queue,
+        persistence_queue=persistence_queue,
+    )
+
+    async def _run() -> None:
+        await orchestrator.run()
+
+    with caplog.at_level(logging.ERROR, logger="mcp_plex.loader.orchestrator"):
+        with pytest.raises(RuntimeError, match="ingestion boom"):
+            asyncio.run(_run())
+
+    assert enrichment_stage.cancelled.is_set()
+    assert persistence_stage.cancelled.is_set()
+    assert ingest_queue.qsize() == 0
+    assert persistence_queue.qsize() == 0
+    error_messages = [record.getMessage() for record in caplog.records]
+    assert any("Ingestion stage failed" in message for message in error_messages)
+
+
+def test_persistence_failure_cancels_upstream(caplog: pytest.LogCaptureFixture) -> None:
+    ingest_queue: asyncio.Queue[object] = asyncio.Queue()
+    persistence_queue: asyncio.Queue[object] = asyncio.Queue()
+    ingestion_stage = BlockingIngestionStage(ingest_queue)
+    enrichment_stage = SingleBatchEnrichmentStage(ingest_queue, persistence_queue)
+    persistence_stage = FailingPersistenceStage(persistence_queue)
+    orchestrator = _build_orchestrator(
+        ingestion_stage=ingestion_stage,
+        enrichment_stage=enrichment_stage,
+        persistence_stage=persistence_stage,
+        ingest_queue=ingest_queue,
+        persistence_queue=persistence_queue,
+    )
+
+    async def _run() -> None:
+        await orchestrator.run()
+
+    with caplog.at_level(logging.ERROR, logger="mcp_plex.loader.orchestrator"):
+        with pytest.raises(RuntimeError, match="persistence boom"):
+            asyncio.run(_run())
+
+    assert ingestion_stage.cancelled.is_set()
+    assert enrichment_stage.cancelled.is_set()
+    assert ingest_queue.qsize() == 0
+    assert persistence_queue.qsize() == 0
+    error_messages = [record.getMessage() for record in caplog.records]
+    assert any("Persistence stage" in message for message in error_messages)
diff --git a/uv.lock b/uv.lock
index 23b6113..9d0386a 100644
--- a/uv.lock
+++ b/uv.lock
@@ -730,7 +730,7 @@ wheels = [
 
 [[package]]
 name = "mcp-plex"
-version = "0.26.70"
+version = "0.26.71"
 source = { editable = "." }
 dependencies = [
     { name = "fastapi" },