feat(experiments/mcp): Event search for shared metrics (#67283)

Author: mp-hog

ee/clickhouse/views/experiment_saved_metrics.py

@@ -1,6 +1,7 @@
+from django.db.models import QuerySet
 from django.db.models.functions import Lower
 
-from drf_spectacular.utils import extend_schema
+from drf_spectacular.utils import OpenApiParameter, extend_schema, extend_schema_view
 from rest_framework import filters, serializers, viewsets
 
 from posthog.api.routing import TeamAndOrgViewSetMixin
@@ -9,7 +10,7 @@
 from posthog.rbac.user_access_control import UserAccessControlSerializerMixin
 
 from products.experiments.backend.experiment_saved_metric_service import ExperimentSavedMetricService
-from products.experiments.backend.metric_utils import refresh_action_names_in_metric
+from products.experiments.backend.metric_utils import filter_metric_group_ids_by_event, refresh_action_names_in_metric
 from products.experiments.backend.models.experiment import ExperimentSavedMetric, ExperimentToSavedMetric
 
 from ee.api.rbac.access_control import AccessControlViewSetMixin
@@ -137,14 +138,50 @@ def _build_service(self) -> ExperimentSavedMetricService:
         return ExperimentSavedMetricService(team=self.context["get_team"](), user=request.user)
 
 
+@extend_schema_view(
+    list=extend_schema(
+        parameters=[
+            OpenApiParameter(
+                name="event",
+                type=str,
+                location=OpenApiParameter.QUERY,
+                required=False,
+                description="Filter to shared metrics whose query references this event name. Matches events "
+                "used directly in metric queries as well as events behind any actions those metrics reference. "
+                "Use this for reuse discovery (find a metric by what it measures); distinct from 'search', "
+                "which matches the metric's own name/description/tags.",
+            ),
+        ],
+    ),
+)
 @extend_schema(extensions={"x-swagger-tag": "experiment_saved_metrics", "x-product": "experiments"})
 class ExperimentSavedMetricViewSet(TeamAndOrgViewSetMixin, AccessControlViewSetMixin, viewsets.ModelViewSet):
     scope_object = "experiment_saved_metric"
     queryset = ExperimentSavedMetric.objects.prefetch_related("created_by").order_by(Lower("name")).distinct()
     serializer_class = ExperimentSavedMetricSerializer
     filter_backends = [filters.SearchFilter]
+    # `search` matches the metric's own name/description/tags, while `event` looks in metrics
     search_fields = ["name", "description", "tagged_items__tag__name"]
 
+    def safely_get_queryset(self, queryset: QuerySet) -> QuerySet:
+        # `?event=` matches metrics whose query references the event directly (EventsNode) or via an action's
+        # step events (ActionsNode resolved live) — mirrors the experiments-list filter. Scoped to `list` so it
+        # never narrows a detail/update/destroy lookup. Rows are read team-scoped (safely_get_queryset runs
+        # before the mixin's team filter), then matched in Python since event references live deep in the JSON.
+        if self.action != "list":
+            return queryset
+        event = self.request.query_params.get("event")
+        if event:
+            # One group per saved metric: (pk, [its query]). `.order_by()` drops the class-level
+            # ordering for this internal fetch — the result is matched in Python, so ordering is
+            # irrelevant, and it avoids a needless sort (and the DISTINCT+ORDER BY column injection).
+            groups = [
+                (pk, [query] if query else [])
+                for pk, query in queryset.filter(team_id=self.team.pk).order_by().values_list("pk", "query")
+            ]
+            queryset = queryset.filter(pk__in=filter_metric_group_ids_by_event(groups, event, self.team))
+        return queryset
+
     def perform_destroy(self, instance: ExperimentSavedMetric) -> None:
         service = ExperimentSavedMetricService(team=self.team, user=self.request.user)
         service.delete_saved_metric(instance)

ee/clickhouse/views/test/test_experiment_saved_metrics.py

@@ -3,8 +3,9 @@
 from parameterized import parameterized
 from rest_framework import status
 
+from products.actions.backend.models.action import Action
 from products.experiments.backend.experiment_saved_metric_service import ExperimentSavedMetricService
-from products.experiments.backend.models.experiment import Experiment, ExperimentToSavedMetric
+from products.experiments.backend.models.experiment import Experiment, ExperimentSavedMetric, ExperimentToSavedMetric
 
 from ee.api.test.base import APILicensedTest
 
@@ -981,3 +982,73 @@ def test_list_paginates_with_limit_and_offset(self) -> None:
         ids_page1 = {row["id"] for row in page1["results"]}
         ids_page2 = {row["id"] for row in page2["results"]}
         self.assertEqual(ids_page1 & ids_page2, set())
+
+    def _seed_metrics_for_event_filter(self) -> None:
+        # Created via the ORM so the stored query JSON is exactly what the filter inspects. The action fires
+        # on the "signup" event, so its metric is discoverable by that event — not by the action's label.
+        signup_action = Action.objects.create(team=self.team, name="Completed signup", steps_json=[{"event": "signup"}])
+        for name, source in [
+            ("Pageview mean", {"kind": "EventsNode", "event": "$pageview"}),
+            ("Purchase mean", {"kind": "EventsNode", "event": "purchase"}),
+            ("Signup via action", {"kind": "ActionsNode", "id": signup_action.id, "name": "Completed signup"}),
+        ]:
+            ExperimentSavedMetric.objects.create(
+                team=self.team,
+                created_by=self.user,
+                name=name,
+                query={"kind": "ExperimentMetric", "metric_type": "mean", "source": source},
+            )
+
+    @parameterized.expand(
+        [
+            # `event` matches the events a metric references — directly, or via an action's step events.
+            ("direct_event", "$pageview", {"Pageview mean"}),
+            ("another_direct_event", "purchase", {"Purchase mean"}),
+            ("event_behind_an_action", "signup", {"Signup via action"}),
+            # It matches events, not the action's label, nor query structure/type tokens.
+            ("action_label_not_matched", "Completed signup", set()),
+            ("metric_type_token_not_matched", "mean", set()),
+            ("node_kind_token_not_matched", "EventsNode", set()),
+            ("no_match", "not_an_event", set()),
+        ]
+    )
+    def test_event_filter_matches_referenced_events_only(
+        self, _name: str, event: str, expected_names: set[str]
+    ) -> None:
+        self._seed_metrics_for_event_filter()
+
+        response = self.client.get(f"/api/projects/{self.team.id}/experiment_saved_metrics/", data={"event": event})
+
+        self.assertEqual(response.status_code, status.HTTP_200_OK)
+        returned = {row["name"] for row in response.json()["results"]}
+        self.assertEqual(returned, expected_names)
+
+    def test_event_filter_composes_with_search(self) -> None:
+        # `event` (references) and `search` (name/description/tags) apply through different mechanisms;
+        # both must narrow the result set together (AND), not clobber each other.
+        purchase_query = {
+            "kind": "ExperimentMetric",
+            "metric_type": "mean",
+            "source": {"kind": "EventsNode", "event": "purchase"},
+        }
+        for name in ["Alpha purchase", "Beta purchase"]:
+            ExperimentSavedMetric.objects.create(team=self.team, created_by=self.user, name=name, query=purchase_query)
+
+        response = self.client.get(
+            f"/api/projects/{self.team.id}/experiment_saved_metrics/?event=purchase&search=Alpha"
+        )
+
+        self.assertEqual(response.status_code, status.HTTP_200_OK)
+        returned = {row["name"] for row in response.json()["results"]}
+        self.assertEqual(returned, {"Alpha purchase"})
+
+    def test_event_param_does_not_filter_detail_retrieve(self) -> None:
+        # `event` is a list-only filter; it must never narrow a detail lookup into a 404.
+        metric_id = self._create_saved_metric("Pageview mean")
+
+        response = self.client.get(
+            f"/api/projects/{self.team.id}/experiment_saved_metrics/{metric_id}?event=not_an_event"
+        )
+
+        self.assertEqual(response.status_code, status.HTTP_200_OK)
+        self.assertEqual(response.json()["id"], metric_id)

products/experiments/backend/experiment_service.py

@@ -43,7 +43,7 @@
 from products.experiments.backend.hogql_queries.base_query_utils import is_threshold_supported_math
 from products.experiments.backend.hogql_queries.experiment_metric_fingerprint import compute_metric_fingerprint
 from products.experiments.backend.hogql_queries.funnel_validation import FunnelDWValidator
-from products.experiments.backend.metric_utils import collect_metric_events_and_action_ids, resolve_action_events
+from products.experiments.backend.metric_utils import filter_metric_group_ids_by_event
 from products.experiments.backend.models.experiment import (
     LEGACY_METRIC_KINDS,
     Experiment,
@@ -2658,28 +2658,14 @@ def _experiments_matching_event(self, queryset: QuerySet[Experiment], event: str
             if query:
                 saved_queries_by_experiment[experiment_id].append(query)
 
-        per_experiment: list[tuple[int, set[str], set[int]]] = []
-        all_action_ids: set[int] = set()
-        for pk, metrics, metrics_secondary in inline_metrics:
-            combined: list[dict[str, Any]] = [
-                *(metrics or []),
-                *(metrics_secondary or []),
-                *saved_queries_by_experiment.get(pk, []),
-            ]
-            events, action_ids = collect_metric_events_and_action_ids(combined)
-            per_experiment.append((pk, events, action_ids))
-            all_action_ids |= action_ids
-
-        action_events = resolve_action_events(all_action_ids, self.team)
-
-        matching_ids: list[int] = []
-        for pk, events, action_ids in per_experiment:
-            resolved = set(events)
-            for action_id in action_ids:
-                resolved |= action_events.get(action_id, set())
-            if event in resolved:
-                matching_ids.append(pk)
-        return matching_ids
+        metric_groups: list[tuple[int, list[dict[str, Any]]]] = [
+            (
+                pk,
+                [*(metrics or []), *(metrics_secondary or []), *saved_queries_by_experiment.get(pk, [])],
+            )
+            for pk, metrics, metrics_secondary in inline_metrics
+        ]
+        return filter_metric_group_ids_by_event(metric_groups, event, self.team)
 
     def filter_experiments_queryset(
         self,

products/experiments/backend/metric_utils.py

@@ -202,3 +202,32 @@ def resolve_action_events(action_ids: set[int], team: Team) -> dict[int, set[str
         action.id: {event for event in action.get_step_events() if event}
         for action in Action.objects.filter(id__in=action_ids, team=team, deleted=False)
     }
+
+
+def filter_metric_group_ids_by_event(
+    metric_groups: list[tuple[int, list[dict[str, Any]]]], event: str, team: Team
+) -> list[int]:
+    """Return the ids of metric groups whose queries reference ``event``, directly or via an action.
+
+    Each group is ``(id, [metric_query, ...])`` — a single query for one saved metric, or an experiment's
+    inline + secondary + attached saved-metric queries. Every referenced action is resolved to its step
+    events in one batched query across all groups, so an action-based metric is matched by the events its
+    action fires on (not by the action's stored — possibly stale — name).
+    """
+    per_group: list[tuple[int, set[str], set[int]]] = []
+    all_action_ids: set[int] = set()
+    for group_id, metrics in metric_groups:
+        events, action_ids = collect_metric_events_and_action_ids(metrics)
+        per_group.append((group_id, events, action_ids))
+        all_action_ids |= action_ids
+
+    action_events = resolve_action_events(all_action_ids, team)
+
+    matching_ids: list[int] = []
+    for group_id, events, action_ids in per_group:
+        resolved = set(events)
+        for action_id in action_ids:
+            resolved |= action_events.get(action_id, set())
+        if event in resolved:
+            matching_ids.append(group_id)
+    return matching_ids

products/experiments/frontend/generated/api.schemas.ts

@@ -712,13 +712,23 @@ tools:
         title: List shared experiment metrics
         description: >
             List shared metrics in the current project. Returns paginated results with id, name, description, query,
-            created_at, updated_at, tags. Use experiment-saved-metrics-retrieve for a single metric by ID, or this
-            listing to resolve by name.
+            created_at, updated_at, tags.
+
+
+            To REUSE a metric by what it measures (rather than by its name): pass ?event=<the event you're measuring> to
+            find metrics that reference it — directly or via an action — then compare each candidate's 'query'
+            (metric_type + math) against the metric you'd otherwise build. Use 'search' (name/description/tags) only
+            when the user named a specific metric to attach.
         param_overrides:
             limit:
                 cast: string-int
             offset:
                 cast: string-int
+            event:
+                description: >
+                    Filter to shared metrics whose query references this event name — matched directly (an EventsNode)
+                    or via the step events of any action the metric references. For finding a reusable metric by what it
+                    measures; then confirm the match against each row's 'query'.
         response:
             include:
                 - id

products/experiments/mcp/tools.yaml

@@ -81,16 +81,18 @@ compare on what each metric measures (its `query`), never on its title.
    (mean / funnel / ratio / retention) before searching — see Step 2 to confirm the event exists via
    `read-data-schema`. You can only recognize a duplicate once you know the concrete event/action,
    so this check runs _after_ you've pinned down the event, not before.
-2. **List the library and compare each candidate's `query`.** Call `experiment-saved-metrics-list`
-   and inspect every result's **`query`** field (not just `name`/`description`). A saved metric is a
-   reuse match when its `query` measures the **same event or action with the same `metric_type`**
-   (and compatible `math`) as the metric you'd otherwise build — even if its name is
-   different.
-   - **Match locally, not via `search`.** `search` matches only `name` / `description` / tags —
-     never the underlying event or action — so it cannot find a definition match, and an empty
-     result means nothing here. Page through the full library with `limit`/`offset` and compare each
-     row's `query` yourself. (Use `search` only when the user names a specific saved metric to
-     attach — that's name resolution, not a definition match.)
+2. **Search by the event, then compare each candidate's `query`.** Call `experiment-saved-metrics-list`
+   with `?event=<the event you're measuring>` to find metrics that reference it — matched directly (an
+   `EventsNode`) **or** via the step events of any action a metric references, so action-based metrics are
+   found by the event their action fires on. Then for each returned row, inspect its **`query`** (not the
+   `name`/`description`): a saved metric is a reuse match when its `query` measures the **same event or
+   action with the same `metric_type`** (and compatible `math`) as the metric you'd otherwise build, even
+   if its name is different.
+   - **Match on the event, not the action's name.** An action-based metric is discoverable by the event
+     the action fires on — pass that event, not the action's label.
+   - **Do not use `search` for this.** `search` matches only the metric's own `name` / `description` / tags —
+     never the underlying event or action — so it cannot find a definition match. Use `search` only when the
+     user names a specific saved metric to attach (name resolution, not a definition match).
 3. **If a saved metric matches the definition** — confirm the match with the user by name/description,
    then attach it instead of building a new one:
    - Call `experiment-get` to read the experiment's current `saved_metrics`.

products/experiments/skills/configuring-experiment-analytics/SKILL.md

@@ -3297,7 +3297,7 @@
         }
     },
     "experiment-saved-metrics-list": {
-        "description": "List shared metrics in the current project. Returns paginated results with id, name, description, query, created_at, updated_at, tags. Use experiment-saved-metrics-retrieve for a single metric by ID, or this listing to resolve by name.",
+        "description": "List shared metrics in the current project. Returns paginated results with id, name, description, query, created_at, updated_at, tags.\n\nTo REUSE a metric by what it measures (rather than by its name): pass ?event=<the event you're measuring> to find metrics that reference it — directly or via an action — then compare each candidate's 'query' (metric_type + math) against the metric you'd otherwise build. Use 'search' (name/description/tags) only when the user named a specific metric to attach.",
         "category": "Experiments",
         "feature": "experiments",
         "summary": "List shared experiment metrics",