fix(webhooks): wire receiver to vector sync queue and fix registered URI

The /webhooks/nextcloud endpoint was a no-op stub that logged the
payload and returned 200 OK; webhook deletions never reached Qdrant.
Compounding that, _get_webhook_uri() registered the docker-compose
internal hostname (http://mcp:8000) with Nextcloud whenever
/.dockerenv existed — including ECS Fargate — so cloud deployments
were registering a URL NC could not resolve.

- New vector/webhook_parser.py extracts a DocumentTask from
  NodeCreatedEvent / NodeWrittenEvent / BeforeNodeDeletedEvent
  payloads scoped to */files/Notes/*.md (matching the registered
  preset filters).
- New vector/webhook_receiver.py pushes that task onto the same
  send-stream the scanner uses (app.state.document_send_stream),
  with 503 when sync is not running so NC retries delivery.
- _get_webhook_uri() now prefers NEXTCLOUD_MCP_SERVER_URL over the
  /.dockerenv branch, so the explicit public URL set on cloud tasks
  wins; docker-compose dev still falls back to the internal name when
  no public URL is configured.

Calendar / Tables event parsing is intentionally out of scope here.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
Chris Coutinho
2026-04-30 03:13:50 +02:00
co-authored by Claude Opus 4.7
parent 111ae988f9
commit 2e2a098bee
7 changed files with 624 additions and 48 deletions
+5 -26
View File
@@ -130,6 +130,7 @@ from nextcloud_mcp_server.vector.oauth_sync import (
user_manager_task,
)
from nextcloud_mcp_server.vector.qdrant_client import get_qdrant_client
from nextcloud_mcp_server.vector.webhook_receiver import handle_nextcloud_webhook
logger = logging.getLogger(__name__)
HTTPXClientInstrumentor().instrument()
@@ -1950,30 +1951,6 @@ def get_app(transport: str = "streamable-http", enabled_apps: list[str] | None =
status_code=status_code,
)
async def handle_nextcloud_webhook(request):
"""Test webhook endpoint to capture and log Nextcloud webhook payloads.
This is a temporary endpoint for testing webhook schemas and payloads.
It logs the full payload and returns 200 OK immediately.
"""
try:
payload = await request.json()
logger.info("=" * 80)
logger.info("🔔 Webhook received from Nextcloud:")
logger.info(json.dumps(payload, indent=2, sort_keys=True))
logger.info("=" * 80)
return JSONResponse(
{"status": "received", "timestamp": payload.get("time")},
status_code=200,
)
except Exception as e:
logger.error(f"❌ Failed to parse webhook payload: {e}")
return JSONResponse(
{"error": "invalid_payload", "message": str(e)}, status_code=400
)
# Add Protected Resource Metadata (PRM) endpoint for OAuth mode
routes = []
@@ -1982,11 +1959,13 @@ def get_app(transport: str = "streamable-http", enabled_apps: list[str] | None =
routes.append(Route("/health/ready", health_ready, methods=["GET"]))
logger.info("Health check endpoints enabled: /health/live, /health/ready")
# Add test webhook endpoint (for development/testing)
# Add Nextcloud webhook receiver (queues DocumentTasks for vector sync).
# Implementation lives in vector/webhook_receiver.py; the handler reads
# the send-stream from request.app.state.document_send_stream.
routes.append(
Route("/webhooks/nextcloud", handle_nextcloud_webhook, methods=["POST"])
)
logger.info("Test webhook endpoint enabled: /webhooks/nextcloud")
logger.info("Webhook endpoint enabled: /webhooks/nextcloud")
# Add management API endpoints for Nextcloud PHP app
# Tier 1: Public endpoints (no auth required)
+24 -22
View File
@@ -77,33 +77,37 @@ async def _get_installed_apps(http_client: httpx.AsyncClient) -> list[str]:
def _get_webhook_uri() -> str:
"""Get the webhook endpoint URI for this MCP server.
This function determines the correct webhook URL based on the environment:
1. Uses WEBHOOK_INTERNAL_URL if explicitly set (highest priority)
2. Detects Docker environment and uses internal service name
3. Falls back to NEXTCLOUD_MCP_SERVER_URL
Priority (highest first):
1. ``WEBHOOK_INTERNAL_URL`` — explicit override, e.g. for split
internal/external URLs.
2. ``NEXTCLOUD_MCP_SERVER_URL`` — the configured public URL. This is
set on cloud deployments (ECS, k8s) and is the URL Nextcloud must
POST to.
3. ``/.dockerenv`` (or podman / ``DOCKER_CONTAINER=true``) → internal
docker-compose service name. Only relevant when no public URL is
configured — i.e. local dev where MCP and NC share a Docker
network.
4. ``http://localhost:8000`` — last-resort fallback.
In Docker environments, Nextcloud needs to reach the MCP service using
the internal Docker network hostname (e.g., http://mcp:8000), not localhost.
Returns:
Full webhook endpoint URL accessible from Nextcloud
Note: ECS Fargate containers also expose ``/.dockerenv``. Without this
priority order, cloud deployments would silently register an internal
docker-compose hostname (e.g. ``http://mcp:8000``) that NC cannot
resolve, dropping every webhook delivery.
"""
# Explicit override (highest priority)
webhook_url = os.getenv("WEBHOOK_INTERNAL_URL")
if webhook_url:
return f"{webhook_url}/webhooks/nextcloud"
# Detect Docker environment
# Check for common Docker indicators
is_docker = (
os.path.exists("/.dockerenv") # Docker container marker file
or os.path.exists("/run/.containerenv") # Podman marker
or os.getenv("DOCKER_CONTAINER") == "true" # Explicit flag
)
server_url = os.getenv("NEXTCLOUD_MCP_SERVER_URL")
if server_url:
return f"{server_url}/webhooks/nextcloud"
is_docker = (
os.path.exists("/.dockerenv")
or os.path.exists("/run/.containerenv")
or os.getenv("DOCKER_CONTAINER") == "true"
)
if is_docker:
# In Docker, use internal service name from NEXTCLOUD_MCP_SERVICE_NAME
# or default to 'mcp' (docker-compose service name)
service_name = os.getenv("NEXTCLOUD_MCP_SERVICE_NAME", "mcp")
port = os.getenv("NEXTCLOUD_MCP_PORT", "8000")
logger.debug(
@@ -111,9 +115,7 @@ def _get_webhook_uri() -> str:
)
return f"http://{service_name}:{port}/webhooks/nextcloud"
# Fallback to configured server URL (for non-Docker deployments)
server_url = os.getenv("NEXTCLOUD_MCP_SERVER_URL", "http://localhost:8000")
return f"{server_url}/webhooks/nextcloud"
return "http://localhost:8000/webhooks/nextcloud"
async def _get_authenticated_client(request: Request) -> httpx.AsyncClient:
@@ -0,0 +1,86 @@
"""Parse Nextcloud webhook payloads into DocumentTask objects.
Maps Nextcloud webhook events to vector-sync DocumentTasks. The handler at
``/webhooks/nextcloud`` calls :func:`extract_document_task` and forwards any
non-None result to the same processor send-stream the scanner uses.
Currently scoped to file (note) events. Calendar / Tables events fall through
to ``None`` for now; those parsers can be added in follow-up changes.
See ADR-010 for the design and ``webhook-testing-findings.md`` for real
captured payloads.
"""
import logging
import re
from nextcloud_mcp_server.vector.scanner import DocumentTask
logger = logging.getLogger(__name__)
_FILE_EVENT_CREATED = "OCP\\Files\\Events\\Node\\NodeCreatedEvent"
_FILE_EVENT_WRITTEN = "OCP\\Files\\Events\\Node\\NodeWrittenEvent"
_FILE_EVENT_BEFORE_DELETED = "OCP\\Files\\Events\\Node\\BeforeNodeDeletedEvent"
# Matches paths inside any user's Notes folder ending in .md, e.g.
# "/admin/files/Notes/Sub/Note.md" or "/alice/files/Notes/foo.md".
_NOTES_PATH_RE = re.compile(r"^/[^/]+/files/Notes/.+\.md$")
def extract_document_task(payload: dict) -> DocumentTask | None:
"""Convert a Nextcloud webhook payload into a DocumentTask.
Returns None for any event we don't (yet) handle, or any event whose
target isn't a markdown file under a user's Notes folder. Callers should
treat None as "ignored" — not an error.
"""
try:
event = payload["event"]
event_class = event["class"]
user_id = payload["user"]["uid"]
except (KeyError, TypeError):
logger.debug("Webhook payload missing user/event/class fields")
return None
if event_class in (
_FILE_EVENT_CREATED,
_FILE_EVENT_WRITTEN,
_FILE_EVENT_BEFORE_DELETED,
):
return _parse_file_event(event_class, event, user_id, payload.get("time", 0))
logger.debug("Ignoring webhook for unsupported event: %s", event_class)
return None
def _parse_file_event(
event_class: str, event: dict, user_id: str, time: int
) -> DocumentTask | None:
node = event.get("node") or {}
path = node.get("path", "")
node_id = node.get("id")
if not _NOTES_PATH_RE.match(path):
# Not a note file — could be a parent folder, an unrelated file, etc.
return None
if node_id is None:
# BeforeNodeDeletedEvent should still carry node.id; if it doesn't
# we can't address the Qdrant points to delete. Skip rather than
# guess — the polling scanner will catch up via its grace period.
logger.warning(
"Webhook %s for note %s missing node.id; falling back to scanner",
event_class,
path,
)
return None
operation = "delete" if event_class == _FILE_EVENT_BEFORE_DELETED else "index"
return DocumentTask(
user_id=user_id,
doc_id=str(node_id),
doc_type="note",
operation=operation,
modified_at=int(time),
)
@@ -0,0 +1,82 @@
"""HTTP receiver for Nextcloud webhooks.
Routes inbound webhooks to the same processor send-stream the scanner uses.
The receiver is registered as a Starlette route at ``/webhooks/nextcloud``
in :mod:`nextcloud_mcp_server.app`.
"""
import logging
from starlette.requests import Request
from starlette.responses import JSONResponse
from nextcloud_mcp_server.vector.webhook_parser import extract_document_task
logger = logging.getLogger(__name__)
async def handle_nextcloud_webhook(request: Request) -> JSONResponse:
"""Receive a Nextcloud webhook and queue a DocumentTask for vector sync.
Returns quickly so NC's webhook worker is not blocked. The send-stream is
read from ``request.app.state.document_send_stream``; when vector sync
isn't running we return 503 so NC retries delivery.
"""
try:
payload = await request.json()
except Exception as e:
logger.warning(f"Webhook payload was not valid JSON: {e}")
return JSONResponse(
{"status": "error", "message": "invalid JSON"},
status_code=400,
)
task = extract_document_task(payload)
if task is None:
event_class = (payload.get("event") or {}).get("class", "<missing>")
logger.debug("Webhook ignored (unsupported event): %s", event_class)
return JSONResponse(
{"status": "ignored", "reason": "unsupported event"},
status_code=200,
)
send_stream = getattr(request.app.state, "document_send_stream", None)
if send_stream is None:
logger.warning(
"Webhook received but vector sync is not running; rejecting so NC retries"
)
return JSONResponse(
{"status": "unavailable", "reason": "vector sync not running"},
status_code=503,
)
try:
await send_stream.send(task)
except Exception as e:
logger.error(
"Failed to queue webhook task for %s_%s: %s",
task.doc_type,
task.doc_id,
e,
)
return JSONResponse(
{"status": "error", "message": "queue unavailable"},
status_code=500,
)
logger.info(
"Webhook queued %s_%s (%s) for user %s",
task.doc_type,
task.doc_id,
task.operation,
task.user_id,
)
return JSONResponse(
{
"status": "queued",
"doc_type": task.doc_type,
"doc_id": task.doc_id,
"operation": task.operation,
},
status_code=200,
)
+144
View File
@@ -0,0 +1,144 @@
"""Unit tests for the ``/webhooks/nextcloud`` HTTP receiver.
Builds a minimal Starlette app around ``handle_nextcloud_webhook`` so we can
drive it with ``TestClient`` without standing up the full FastMCP server.
"""
import anyio
import pytest
from starlette.applications import Starlette
from starlette.routing import Route
from starlette.testclient import TestClient
from nextcloud_mcp_server.vector.webhook_receiver import handle_nextcloud_webhook
pytestmark = pytest.mark.unit
def _make_app(send_stream=None) -> Starlette:
app = Starlette(
routes=[
Route("/webhooks/nextcloud", handle_nextcloud_webhook, methods=["POST"])
]
)
app.state.document_send_stream = send_stream
return app
_NOTE_CREATED = {
"user": {"uid": "admin", "displayName": "admin"},
"time": 1762850245,
"event": {
"class": "OCP\\Files\\Events\\Node\\NodeCreatedEvent",
"node": {
"id": 437,
"path": "/admin/files/Notes/Webhooks/Webhook Test Note.md",
},
},
}
_NOTE_DELETED = {
"user": {"uid": "alice"},
"time": 1762851093,
"event": {
"class": "OCP\\Files\\Events\\Node\\BeforeNodeDeletedEvent",
"node": {"id": 99, "path": "/alice/files/Notes/foo.md"},
},
}
def test_index_event_queues_task_and_returns_200():
send_stream, receive_stream = anyio.create_memory_object_stream(max_buffer_size=4)
app = _make_app(send_stream=send_stream)
with TestClient(app) as client:
response = client.post("/webhooks/nextcloud", json=_NOTE_CREATED)
assert response.status_code == 200
assert response.json()["status"] == "queued"
assert response.json()["operation"] == "index"
assert response.json()["doc_id"] == "437"
task = receive_stream.receive_nowait()
assert task.user_id == "admin"
assert task.doc_id == "437"
assert task.operation == "index"
assert task.doc_type == "note"
def test_delete_event_queues_delete_task():
send_stream, receive_stream = anyio.create_memory_object_stream(max_buffer_size=4)
app = _make_app(send_stream=send_stream)
with TestClient(app) as client:
response = client.post("/webhooks/nextcloud", json=_NOTE_DELETED)
assert response.status_code == 200
assert response.json()["operation"] == "delete"
task = receive_stream.receive_nowait()
assert task.operation == "delete"
assert task.doc_id == "99"
assert task.user_id == "alice"
def test_unsupported_event_is_ignored():
send_stream, receive_stream = anyio.create_memory_object_stream(max_buffer_size=4)
app = _make_app(send_stream=send_stream)
payload = {
"user": {"uid": "admin"},
"time": 1,
"event": {
"class": "OCP\\Calendar\\Events\\CalendarObjectCreatedEvent",
"objectData": {"id": 7},
},
}
with TestClient(app) as client:
response = client.post("/webhooks/nextcloud", json=payload)
assert response.status_code == 200
assert response.json()["status"] == "ignored"
with pytest.raises(anyio.WouldBlock):
receive_stream.receive_nowait()
def test_invalid_json_returns_400():
app = _make_app(send_stream=None)
with TestClient(app) as client:
response = client.post(
"/webhooks/nextcloud",
content=b"not json",
headers={"content-type": "application/json"},
)
assert response.status_code == 400
assert response.json()["status"] == "error"
def test_returns_503_when_send_stream_not_wired():
"""Vector sync not running → tell NC to retry instead of dropping the
event."""
app = _make_app(send_stream=None)
with TestClient(app) as client:
response = client.post("/webhooks/nextcloud", json=_NOTE_CREATED)
assert response.status_code == 503
assert response.json()["status"] == "unavailable"
def test_returns_500_when_stream_is_closed():
send_stream, receive_stream = anyio.create_memory_object_stream(max_buffer_size=1)
receive_stream.close() # close receiver → send raises BrokenResourceError
app = _make_app(send_stream=send_stream)
with TestClient(app) as client:
response = client.post("/webhooks/nextcloud", json=_NOTE_CREATED)
assert response.status_code == 500
assert response.json()["status"] == "error"
+188
View File
@@ -0,0 +1,188 @@
"""Unit tests for the Nextcloud webhook payload parser.
Payload examples are taken from real Nextcloud captures recorded in
``webhook-testing-findings.md``.
"""
import pytest
from nextcloud_mcp_server.vector.webhook_parser import extract_document_task
@pytest.mark.unit
def test_node_created_event_returns_index_task():
payload = {
"user": {"uid": "admin", "displayName": "admin"},
"time": 1762850245,
"event": {
"class": "OCP\\Files\\Events\\Node\\NodeCreatedEvent",
"node": {
"id": 437,
"path": "/admin/files/Notes/Webhooks/Webhook Test Note.md",
},
},
}
task = extract_document_task(payload)
assert task is not None
assert task.user_id == "admin"
assert task.doc_id == "437"
assert task.doc_type == "note"
assert task.operation == "index"
assert task.modified_at == 1762850245
@pytest.mark.unit
def test_node_written_event_returns_index_task():
payload = {
"user": {"uid": "admin", "displayName": "admin"},
"time": 1762850960,
"event": {
"class": "OCP\\Files\\Events\\Node\\NodeWrittenEvent",
"node": {
"id": 437,
"path": "/admin/files/Notes/Webhooks/Webhook Test Note.md",
},
},
}
task = extract_document_task(payload)
assert task is not None
assert task.operation == "index"
assert task.doc_id == "437"
@pytest.mark.unit
def test_before_node_deleted_event_returns_delete_task():
payload = {
"user": {"uid": "alice", "displayName": "Alice"},
"time": 1762851093,
"event": {
"class": "OCP\\Files\\Events\\Node\\BeforeNodeDeletedEvent",
"node": {
"id": 437,
"path": "/alice/files/Notes/Webhooks/Webhook Test Note.md",
},
},
}
task = extract_document_task(payload)
assert task is not None
assert task.user_id == "alice"
assert task.operation == "delete"
assert task.doc_id == "437"
assert task.doc_type == "note"
@pytest.mark.unit
def test_node_id_is_normalized_to_string():
"""NC sends node.id as int; we always emit str (per ADR-010 §A.3)."""
payload = {
"user": {"uid": "admin"},
"time": 1,
"event": {
"class": "OCP\\Files\\Events\\Node\\NodeCreatedEvent",
"node": {"id": 437, "path": "/admin/files/Notes/foo.md"},
},
}
task = extract_document_task(payload)
assert task is not None
assert isinstance(task.doc_id, str)
assert task.doc_id == "437"
@pytest.mark.unit
def test_path_outside_notes_returns_none():
payload = {
"user": {"uid": "admin"},
"time": 1,
"event": {
"class": "OCP\\Files\\Events\\Node\\NodeCreatedEvent",
"node": {"id": 1, "path": "/admin/files/Documents/foo.md"},
},
}
assert extract_document_task(payload) is None
@pytest.mark.unit
def test_non_markdown_inside_notes_returns_none():
payload = {
"user": {"uid": "admin"},
"time": 1,
"event": {
"class": "OCP\\Files\\Events\\Node\\NodeCreatedEvent",
"node": {"id": 1, "path": "/admin/files/Notes/image.png"},
},
}
assert extract_document_task(payload) is None
@pytest.mark.unit
def test_parent_folder_event_returns_none():
"""Creating a note fires events for the parent folder too — ignore those."""
payload = {
"user": {"uid": "admin"},
"time": 1,
"event": {
"class": "OCP\\Files\\Events\\Node\\NodeCreatedEvent",
"node": {"id": 100, "path": "/admin/files/Notes/Webhooks"},
},
}
assert extract_document_task(payload) is None
@pytest.mark.unit
def test_unknown_event_class_returns_none():
payload = {
"user": {"uid": "admin"},
"time": 1,
"event": {
"class": "OCP\\Calendar\\Events\\CalendarObjectCreatedEvent",
"objectData": {"id": 7, "uri": "x.ics"},
},
}
assert extract_document_task(payload) is None
@pytest.mark.unit
def test_node_deleted_event_without_id_returns_none():
"""``NodeDeletedEvent`` (no node.id) is not the registered event, but if
one ever leaks through we ignore it rather than guess at the doc_id —
the polling scanner will catch up via its grace period."""
payload = {
"user": {"uid": "admin"},
"time": 1,
"event": {
"class": "OCP\\Files\\Events\\Node\\BeforeNodeDeletedEvent",
"node": {"path": "/admin/files/Notes/foo.md"},
},
}
assert extract_document_task(payload) is None
@pytest.mark.unit
def test_missing_user_field_returns_none():
payload = {
"time": 1,
"event": {
"class": "OCP\\Files\\Events\\Node\\NodeCreatedEvent",
"node": {"id": 1, "path": "/admin/files/Notes/foo.md"},
},
}
assert extract_document_task(payload) is None
@pytest.mark.unit
def test_empty_payload_returns_none():
assert extract_document_task({}) is None
+95
View File
@@ -0,0 +1,95 @@
"""Unit tests for ``_get_webhook_uri`` priority order.
Cloud deployments register the webhook URI returned by this function with
Nextcloud. ECS Fargate also exposes ``/.dockerenv``, so an explicit public
URL must win over the docker auto-detection branch.
"""
import pytest
from nextcloud_mcp_server.auth.webhook_routes import _get_webhook_uri
ENV_VARS = (
"WEBHOOK_INTERNAL_URL",
"NEXTCLOUD_MCP_SERVER_URL",
"NEXTCLOUD_MCP_SERVICE_NAME",
"NEXTCLOUD_MCP_PORT",
"DOCKER_CONTAINER",
)
@pytest.fixture(autouse=True)
def _clean_env(monkeypatch):
for name in ENV_VARS:
monkeypatch.delenv(name, raising=False)
def _no_docker_markers(monkeypatch):
monkeypatch.setattr(
"nextcloud_mcp_server.auth.webhook_routes.os.path.exists",
lambda _path: False,
)
def _docker_markers(monkeypatch):
monkeypatch.setattr(
"nextcloud_mcp_server.auth.webhook_routes.os.path.exists",
lambda path: path == "/.dockerenv",
)
@pytest.mark.unit
def test_webhook_internal_url_wins_over_everything(monkeypatch):
monkeypatch.setenv("WEBHOOK_INTERNAL_URL", "https://internal.example.com")
monkeypatch.setenv("NEXTCLOUD_MCP_SERVER_URL", "https://public.example.com")
_docker_markers(monkeypatch)
assert _get_webhook_uri() == "https://internal.example.com/webhooks/nextcloud"
@pytest.mark.unit
def test_public_url_wins_over_docker_detection(monkeypatch):
"""The bug-fix case: ECS containers have /.dockerenv but a public URL is
set. Docker auto-detection must NOT clobber the explicit public URL."""
monkeypatch.setenv(
"NEXTCLOUD_MCP_SERVER_URL", "https://holy-bluegill.astrolabecloud.com"
)
_docker_markers(monkeypatch)
assert (
_get_webhook_uri()
== "https://holy-bluegill.astrolabecloud.com/webhooks/nextcloud"
)
@pytest.mark.unit
def test_docker_detection_used_when_no_public_url(monkeypatch):
"""docker-compose dev: no public URL set, /.dockerenv exists → use the
docker-compose service name."""
_docker_markers(monkeypatch)
assert _get_webhook_uri() == "http://mcp:8000/webhooks/nextcloud"
@pytest.mark.unit
def test_docker_detection_honors_service_name_and_port_overrides(monkeypatch):
monkeypatch.setenv("NEXTCLOUD_MCP_SERVICE_NAME", "mcp-login-flow")
monkeypatch.setenv("NEXTCLOUD_MCP_PORT", "8004")
_docker_markers(monkeypatch)
assert _get_webhook_uri() == "http://mcp-login-flow:8004/webhooks/nextcloud"
@pytest.mark.unit
def test_docker_container_env_var_triggers_docker_branch(monkeypatch):
monkeypatch.setenv("DOCKER_CONTAINER", "true")
_no_docker_markers(monkeypatch)
assert _get_webhook_uri() == "http://mcp:8000/webhooks/nextcloud"
@pytest.mark.unit
def test_localhost_fallback_when_nothing_set(monkeypatch):
_no_docker_markers(monkeypatch)
assert _get_webhook_uri() == "http://localhost:8000/webhooks/nextcloud"