feat(status-pages): API v2 and MCP tools

apps/ai/src/mcp/dispatcher.ts

@@ -28,6 +28,8 @@ import { RailwayMetricsService } from "@maple/backend/services/integrations/Rail
 import { ScrapeTargetsService } from "@maple/backend/services/integrations/ScrapeTargetsService"
 import { UptimeMonitorsService } from "@maple/backend/services/uptime/UptimeMonitorsService"
 import { IngestRuleService } from "@maple/backend/services/org/IngestRuleService"
+import { StatusPagesService } from "@maple/backend/services/status-pages/StatusPagesService"
+import { StatusPageSnapshotBuilder } from "@maple/backend/services/status-pages/StatusPageSnapshotBuilder"
 import { VcsCommitService } from "@maple/backend/services/integrations/vcs/VcsCommitService"
 import { IntegrationInventoryService } from "@maple/backend/services/integrations/IntegrationInventoryService"
 import { SourcesService } from "@maple/backend/services/org/SourcesService"
@@ -274,6 +276,8 @@ const McpRuntimeServicesLive = Layer.mergeAll(
 	RepoSandboxService.layer,
 	ScrapeTargetsService.layer,
 	SetupAuditService.layer,
+	StatusPageSnapshotBuilder.layer,
+	StatusPagesService.layer,
 	UptimeMonitorsService.layer,
 	IngestRuleService.layer,
 	VcsCommitService.layer,

apps/ai/src/mcp/resources/instructions.ts

@@ -116,6 +116,7 @@ first occurrences ──> (candidate) ──> issue in \`triage\`
 - Deploy verification: service_deployments (versions per service, newest vs previous) -> compare_periods around the rollout. Route usage over up to 90 days: route_usage. Silence check before calling an outage: ingest_freshness. Volume per service: ingest_usage. Database query shapes by volume: db_query_volume
 - Alert management: list_alert_rules / list_alert_destinations -> get_alert_rule -> preview_alert_rule (dry-run before saving; a window with no data is a blind rule, not a healthy one) -> create_alert_rule / update_alert_rule / delete_alert_rule -> list_alert_incidents
 - Uptime: list_uptime_monitors is every http/tcp/dns check and heartbeat with its status and last check; a failed check's trace id goes to inspect_trace. Admins manage them with create_uptime_monitor / update_uptime_monitor (enabled=false pauses) / delete_uptime_monitor. A down monitor opens an alert incident with signal type \`uptime\`; its rule is managed only through the monitor.
+- Status pages (what customers see, separate from alert incidents): list_status_pages -> get_status_page (component statuses, open incidents, freshness) -> create_status_page_incident / schedule_status_page_maintenance (drafts unless publish=true) -> post_status_page_update -> resolve_status_page_incident. Every message is public text.
 - AI agent sessions (LLM agent traces — NOT browser replays): list_agent_sessions -> get_agent_session (verdict and checks) -> get_agent_session_transcript (what it did step by step; \`failed_only\`, \`turn\`) -> inspect_span (one span in full)
 - Agent tool health: get_agent_tools_overview -> get_agent_tools_overview tool="<tool>" (its failure groups) -> get_agent_tool_error
 - Product analytics / conversion: list_product_events -> query_funnel (steps over page views, \`track()\` events and server events, stitched per person; \`breakdown_by\` a UTM/referrer dimension or an event attribute) -> add_dashboard_widget with \`panel_type: "funnel"\` and \`display_json.funnel.steps\` to pin it (\`display_json.funnel.variant: "dropoff"\` for the step-by-step drop-off view), or \`panel_type: "paths"\` with \`display_json.paths.anchor\` for what people do after / before one event

apps/ai/src/mcp/tools/__tests__/status-page-tools.test.ts

@@ -0,0 +1,227 @@
+import { afterAll, beforeAll, describe, expect, it } from "vitest"
+import { Schema } from "effect"
+import {
+	GetStatusPageOutput,
+	ListStatusPagesOutput,
+	StatusPageIncidentMutationOutput,
+} from "@maple/domain/mcp-outputs"
+import { executeSql } from "@maple/backend/platform/test-pglite"
+import type { McpToolResult } from "../types"
+import { makeEvalRuntime, markdown, runToolDirect, type EvalRuntime } from "../../__evals__/eval-runtime"
+
+let rt: EvalRuntime
+beforeAll(async () => {
+	rt = makeEvalRuntime()
+	const now = Date.now()
+	await executeSql(
+		rt.testDb,
+		`INSERT INTO status_pages (id, org_id, name, slug, created_by, created_at, updated_at)
+		 VALUES ($1, $2, 'Acme Status', 'acme-status', $3, to_timestamp($4 / 1000.0), to_timestamp($4 / 1000.0))`,
+		["7f0b7d0e-0b6a-4f55-9a0e-2a4b9a0c1d01", rt.tenant.orgId, rt.tenant.userId, now],
+	)
+	await executeSql(
+		rt.testDb,
+		`INSERT INTO status_page_components (id, org_id, page_id, name, position, created_at, updated_at)
+		 VALUES ($1, $2, $3, 'API', 0, to_timestamp($4 / 1000.0), to_timestamp($4 / 1000.0))`,
+		[
+			"7f0b7d0e-0b6a-4f55-9a0e-2a4b9a0c1d02",
+			rt.tenant.orgId,
+			"7f0b7d0e-0b6a-4f55-9a0e-2a4b9a0c1d01",
+			now,
+		],
+	)
+})
+afterAll(async () => {
+	await rt.dispose()
+})
+
+const decode = <S extends Schema.Codec<unknown, unknown, never, never>>(schema: S, result: McpToolResult) =>
+	Schema.decodeUnknownSync(schema)(result.structuredContent)
+
+describe("status page tools", () => {
+	it("lists the org's pages with their REST ids", async () => {
+		const output = decode(ListStatusPagesOutput, await runToolDirect(rt, "list_status_pages", {}))
+		expect(output.total).toBe(1)
+		expect(output.pages[0]).toMatchObject({ name: "Acme Status", slug: "acme-status" })
+		expect(output.pages[0]?.id).toMatch(/^stp_/)
+	})
+
+	it("opens an agent incident as a draft that the page does not show", async () => {
+		const created = decode(
+			StatusPageIncidentMutationOutput,
+			await runToolDirect(rt, "create_status_page_incident", {
+				page: "acme-status",
+				title: "Elevated API errors",
+				message: "We are looking into elevated error rates.",
+				components: ["API"],
+				impact: "major",
+			}),
+		)
+		expect(created.incident).toMatchObject({
+			kind: "incident",
+			status: "investigating",
+			published: false,
+			impact: "major",
+		})
+		expect(created.incident.id).toMatch(/^sti_/)
+
+		// A draft names the component but does not change what visitors see.
+		const page = decode(
+			GetStatusPageOutput,
+			await runToolDirect(rt, "get_status_page", { page: "acme-status" }),
+		)
+		expect(page.components).toMatchObject([{ name: "API", status: "operational" }])
+		expect(page.activeIncidents.map((i) => i.id)).toContain(created.incident.id)
+		expect(page.freshness.publishPending).toBe(true)
+
+		// Publishing with the next update puts it on the page.
+		const updated = decode(
+			StatusPageIncidentMutationOutput,
+			await runToolDirect(rt, "post_status_page_update", {
+				incident: created.incident.id,
+				status: "identified",
+				message: "We found the cause and are rolling back.",
+				publish: true,
+			}),
+		)
+		expect(updated.incident).toMatchObject({ status: "identified", published: true, updateCount: 2 })
+		const liveResult = await runToolDirect(rt, "get_status_page", { page: "Acme Status" })
+		// The follow-up needs public wording, so it is a notice, not a half-filled call.
+		expect(markdown(liveResult)).toContain("To update an open incident, call post_status_page_update")
+		const live = decode(GetStatusPageOutput, liveResult)
+		expect(live.components).toMatchObject([{ name: "API", status: "partial_outage" }])
+		expect(live.overallStatus).toBe("partial_outage")
+
+		const resolved = decode(
+			StatusPageIncidentMutationOutput,
+			await runToolDirect(rt, "resolve_status_page_incident", {
+				incident: created.incident.id,
+				message: "Error rates are back to normal.",
+			}),
+		)
+		expect(resolved.incident.status).toBe("resolved")
+		expect(resolved.incident.resolvedAt).not.toBeNull()
+	})
+
+	it("publishes straight away when asked", async () => {
+		const created = decode(
+			StatusPageIncidentMutationOutput,
+			await runToolDirect(rt, "create_status_page_incident", {
+				page: "acme-status",
+				title: "Login delays",
+				message: "Some sign-ins are slow.",
+				publish: true,
+			}),
+		)
+		expect(created.incident.published).toBe(true)
+	})
+
+	it("schedules maintenance as a draft with its window", async () => {
+		const result = await runToolDirect(rt, "schedule_status_page_maintenance", {
+			page: "acme-status",
+			title: "Database upgrade",
+			message: "The API may be read-only for a few minutes.",
+			starts_at: "2026-08-01 02:00:00",
+			ends_at: "2026-08-01T03:00:00Z",
+			components: ["API"],
+		})
+		const output = decode(StatusPageIncidentMutationOutput, result)
+		expect(output.incident).toMatchObject({
+			kind: "maintenance",
+			status: "scheduled",
+			published: false,
+			scheduledStart: "2026-08-01T02:00:00.000Z",
+			scheduledEnd: "2026-08-01T03:00:00.000Z",
+		})
+		expect(markdown(result)).toContain("draft")
+	})
+
+	it("names the unknown component and page instead of guessing", async () => {
+		const component = await runToolDirect(rt, "create_status_page_incident", {
+			page: "acme-status",
+			title: "x",
+			message: "y",
+			components: ["Billing"],
+		})
+		expect(component.isError).toBe(true)
+		expect(markdown(component)).toContain("API")
+		const page = await runToolDirect(rt, "get_status_page", { page: "nope" })
+		expect(page.isError).toBe(true)
+		expect(markdown(page)).toContain("list_status_pages")
+	})
+
+	it("resolves a draft and publishes it in the same call, or says it stayed a draft", async () => {
+		const open = async (title: string) =>
+			decode(
+				StatusPageIncidentMutationOutput,
+				await runToolDirect(rt, "create_status_page_incident", {
+					page: "acme-status",
+					title,
+					message: "Investigating.",
+				}),
+			).incident.id
+		const kept = await runToolDirect(rt, "resolve_status_page_incident", {
+			incident: await open("Quiet draft"),
+			message: "Never mind.",
+		})
+		expect(decode(StatusPageIncidentMutationOutput, kept).incident).toMatchObject({
+			status: "resolved",
+			published: false,
+		})
+		expect(markdown(kept)).toContain("closed as a draft")
+		expect(markdown(kept)).not.toContain("post_status_page_update")
+
+		const shown = decode(
+			StatusPageIncidentMutationOutput,
+			await runToolDirect(rt, "resolve_status_page_incident", {
+				incident: await open("Public draft"),
+				message: "Fixed.",
+				publish: true,
+			}),
+		)
+		expect(shown.incident).toMatchObject({ status: "resolved", published: true })
+	})
+
+	it("reads delayed while the page's monitors have not reported, and live once one has", async () => {
+		const monitorId = "7f0b7d0e-0b6a-4f55-9a0e-2a4b9a0c1d09"
+		const now = Date.now()
+		await executeSql(
+			rt.testDb,
+			`INSERT INTO uptime_monitors (id, org_id, name, kind, config_json, created_by, created_at, updated_at, enabled)
+			 VALUES ($1, $2, 'API health', 'http', $3, $4, now(), now(), true)`,
+			[
+				monitorId,
+				rt.tenant.orgId,
+				JSON.stringify({
+					_tag: "http",
+					url: "https://example.com/health",
+					method: "GET",
+					headers: {},
+					assertions: [],
+					timeoutMs: 5000,
+				}),
+				rt.tenant.userId,
+			],
+		)
+		await executeSql(
+			rt.testDb,
+			`INSERT INTO uptime_monitor_states (monitor_id, org_id, status, recent_json, updated_at) VALUES ($1, $2, 'up', '[]', now())`,
+			[monitorId, rt.tenant.orgId],
+		)
+		await executeSql(rt.testDb, `UPDATE status_page_components SET monitor_id = $1 WHERE name = 'API'`, [
+			monitorId,
+		])
+		const freshness = async () =>
+			decode(GetStatusPageOutput, await runToolDirect(rt, "get_status_page", { page: "acme-status" }))
+				.freshness
+
+		expect(await freshness()).toMatchObject({ state: "delayed", updatedAt: null })
+		await executeSql(
+			rt.testDb,
+			`INSERT INTO uptime_monitor_daily (monitor_id, day, org_id, checks, failures, down_checks, last_check_at)
+			 VALUES ($1, $2, $3, 1, 0, 0, to_timestamp($4 / 1000.0))`,
+			[monitorId, new Date(now).toISOString().slice(0, 10), rt.tenant.orgId, now - 30_000],
+		)
+		expect((await freshness()).state).toBe("live")
+	})
+})

apps/ai/src/mcp/tools/mutating.ts

@@ -43,6 +43,11 @@ export const MUTATING_TOOL_NAMES: ReadonlySet<string> = new Set([
 	"create_ingest_rule",
 	"update_ingest_rule",
 	"delete_ingest_rule",
+	// status pages: every post is public text on the customer's page
+	"create_status_page_incident",
+	"post_status_page_update",
+	"resolve_status_page_incident",
+	"schedule_status_page_maintenance",
 	// feedback
 	"send_maple_feedback",
 ])

apps/ai/src/mcp/tools/registry.ts

@@ -27,6 +27,7 @@ import { registerCompareCommitsTool } from "./compare-commits"
 import { registerIntegrationWriteTools } from "./integration-writes"
 import { registerUptimeMonitorTools } from "./uptime-monitors"
 import { registerIngestRuleTools } from "./ingest-rules"
+import { registerStatusPageTools } from "./status-pages"
 import { registerGetInstrumentationRecommendationsTool } from "./get-instrumentation-recommendations"
 import { registerGetServiceTopOperationsTool } from "./get-service-top-operations"
 import { registerServiceDeploymentsTool } from "./service-deployments"
@@ -435,6 +436,7 @@ const collectMapleToolDefinitions = (): ReadonlyArray<MapleToolDefinition> => {
 	registerIntegrationWriteTools(registrar)
 	registerUptimeMonitorTools(registrar)
 	registerIngestRuleTools(registrar)
+	registerStatusPageTools(registrar)
 	registerSourceCodeTools(registrar)
 	registerSandboxTools(registrar)
 	registerPullRequestTools(registrar)

apps/ai/src/mcp/tools/runtime-requirements.ts

@@ -19,6 +19,8 @@ import type { VcsCommitService } from "@maple/backend/services/integrations/vcs/
 import type { VcsSourceService } from "@maple/backend/services/integrations/vcs/VcsSourceService"
 import type { UptimeMonitorsService } from "@maple/backend/services/uptime/UptimeMonitorsService"
 import type { IngestRuleService } from "@maple/backend/services/org/IngestRuleService"
+import type { StatusPagesService } from "@maple/backend/services/status-pages/StatusPagesService"
+import type { StatusPageSnapshotBuilder } from "@maple/backend/services/status-pages/StatusPageSnapshotBuilder"
 import type { SetupAuditService } from "@maple/backend/services/org/SetupAuditService"
 import type { RepoSandboxService } from "@maple/backend/services/sandbox/RepoSandboxService"
 import type { QueryEngineService } from "@maple/backend/services/warehouse/QueryEngineService"
@@ -51,6 +53,8 @@ export type McpToolRuntimeRequirements =
 	| RepoSandboxService
 	| ScrapeTargetsService
 	| SetupAuditService
+	| StatusPageSnapshotBuilder
+	| StatusPagesService
 	| UptimeMonitorsService
 	| IngestRuleService
 	| VcsCommitService