nuxt · HugoRCD · Jul 24, 2026 · Jul 22, 2026 · Jul 22, 2026 · Jul 22, 2026
@@ -47,6 +47,11 @@ INTERNAL_API_SECRET=
 # when pre-rendering the docs)
 NUXT_PUBLIC_SITE_URL=
 
+# nuxt-js team / nuxt project Vercel ids (admin-only Vercel MCP connection).
+# Find them in `.vercel/project.json` (orgId / projectId) after `vercel link`.
+NUXI_VERCEL_PROJECT_ID=
+NUXI_VERCEL_TEAM_ID=
+
 # ----------------------------------------------------------------------------
 # GitHub API data (optional — repo stats, contributors, team members)
 # ----------------------------------------------------------------------------

@@ -26,3 +26,5 @@ dist
 .wrangler
 test-results/
 .env*.local
+.env*
+!.env.example
@@ -104,17 +104,29 @@ export default defineSchedule({
 })
 ```
 
+All times below assume UTC+1 local mornings (6:00 local ≈ 5:00 UTC) — adjust the cron if the team's timezone/DST differs.
+
 ### Weekly digest
 
-- Schedule: `agent/schedules/weekly-digest.ts` — Monday 9:00 UTC
+- Schedule: `agent/schedules/weekly-digest.ts` — Monday 5:00 UTC
 - Skill: `agent/skills/weekly-digest/SKILL.md`
 - Preview trigger: `POST /eve/v1/ops/weekly-digest/trigger`
+- Real traffic pulse (`connection__vercel_mcp__get_web_analytics`), AI Gateway spend/tokens (`ai_gateway__report`), and Slack-vs-web run split (`connection__vercel_mcp__list_agent_runs`) — no more DB-guessed or link-only numbers. **Fix this week** is prioritized using each worst-feedback page's real traffic.
+
+### Analytics digest
+
+Deep traffic dive: trend, top pages/referrers/geo/device (WoW deltas), and doc pages that combine high traffic with poor feedback ("pages that need love").
+
+- Schedule: `agent/schedules/analytics-digest.ts` — Monday 5:15 UTC (right after weekly-digest)
+- Skill: `agent/skills/analytics-digest/SKILL.md`
+- Connection: `connection__vercel_mcp__get_web_analytics` (`agent/connections/vercel-mcp.ts`)
+- Preview trigger: `POST /eve/v1/ops/analytics-digest/trigger?sinceDays=7`
 
 ### Firehose summary
 
 Summarizes `#firehose-nuxt` (Octolens social mentions) and posts highlights to the workflow channel.
 
-- Schedule: `agent/schedules/firehose-summary.ts` — weekdays 9:00 UTC (last 24h)
+- Schedule: `agent/schedules/firehose-summary.ts` — weekdays 5:00 UTC (last 24h)
 - Skill: `agent/skills/firehose-summary/SKILL.md`
 - Tool: `read_slack_channel_history` (`agent/tools/slack-channel-history.ts`)
 - Preview trigger: `POST /eve/v1/ops/firehose-summary/trigger?sinceHours=24`
@@ -131,6 +143,7 @@ With the dev server running (`pnpm dev` from repo root — Eve is bundled via th
 
 ```sh
 curl -X POST "http://localhost:3000/eve/v1/dev/schedules/weekly-digest"
+curl -X POST "http://localhost:3000/eve/v1/dev/schedules/analytics-digest"
 curl -X POST "http://localhost:3000/eve/v1/dev/schedules/firehose-summary"
 # -> { "scheduleId": "...", "sessionIds": ["..."] }
 ```
@@ -141,8 +154,11 @@ curl -X POST "http://localhost:3000/eve/v1/dev/schedules/firehose-summary"
 curl -X POST "https://<preview-url>/eve/v1/ops/weekly-digest/trigger?sinceDays=7" \
   -H "Authorization: Bearer $INTERNAL_API_SECRET"
 
+curl -X POST "https://<preview-url>/eve/v1/ops/analytics-digest/trigger?sinceDays=7" \
+  -H "Authorization: Bearer $INTERNAL_API_SECRET"
+
 curl -X POST "https://<preview-url>/eve/v1/ops/firehose-summary/trigger?sinceHours=24" \
   -H "Authorization: Bearer $INTERNAL_API_SECRET"
 ```
 
-Requires on the **eve** runtime: `INTERNAL_API_SECRET`, `NUXT_MCP_ADMIN_TOKEN`, `NUXT_WORKFLOW_SLACK_CHANNEL`, `NUXT_FIREHOSE_SLACK_CHANNEL` (Slack channel names). Optional `NUXT_*_SLACK_CHANNEL_ID` overrides names and skips `users.conversations`. Local dev and Vercel preview use Connect client `slack/nuxi-preview` automatically; prod uses `slack/nuxi` (override with `SLACK_CONNECTOR`).
+Requires on the **eve** runtime: `INTERNAL_API_SECRET`, `NUXT_MCP_ADMIN_TOKEN`, `NUXT_WORKFLOW_SLACK_CHANNEL`, `NUXT_FIREHOSE_SLACK_CHANNEL` (Slack channel names). Optional `NUXT_*_SLACK_CHANNEL_ID` overrides names and skips `users.conversations`. Local dev and Vercel preview use Connect client `slack/nuxi-preview` automatically; prod uses `slack/nuxi` (override with `SLACK_CONNECTOR`). `weekly-digest` and `analytics-digest` additionally need `NUXI_VERCEL_TEAM_ID`/`NUXI_VERCEL_PROJECT_ID` (see `agent/lib/vercel-connect.ts`) and, for spend/token numbers, `AI_GATEWAY_API_KEY` — both connections are admin-gated so only the scheduled/Slack/admin path can reach them.
@@ -7,6 +7,7 @@ import {
   verifyWorkflowTriggerAuth
 } from '../lib/workflows.js'
 import { runDiscordGateway } from '../schedules/discord-gateway.js'
+import { runAnalyticsDigest } from '../schedules/analytics-digest.js'
 import { runFirehoseSummary } from '../schedules/firehose-summary.js'
 import { runWeeklyDigest } from '../schedules/weekly-digest.js'
 
@@ -66,6 +67,28 @@ export default defineChannel({
       }))
 
       return Response.json({ ok: true, sinceHours: parsedSinceHours.value ?? null })
+    }),
+    POST('/eve/v1/ops/analytics-digest/trigger', async (req, args) => {
+      if (!isManualWorkflowTriggerAllowed()) {
+        return Response.json({ error: 'Manual workflow trigger is disabled' }, { status: 404 })
+      }
+      if (!verifyWorkflowTriggerAuth(req)) {
+        return Response.json({ error: 'Unauthorized' }, { status: 401 })
+      }
+
+      const url = new URL(req.url)
+      const parsedSinceDays = parseSinceDays(url.searchParams.get('sinceDays'))
+      if (!parsedSinceDays.ok) {
+        return Response.json({ error: parsedSinceDays.error }, { status: 400 })
+      }
+
+      args.waitUntil(runAnalyticsDigest({
+        receive: args.receive,
+        appAuth: scheduleAppAuth,
+        sinceDays: parsedSinceDays.value
+      }))
+
+      return Response.json({ ok: true, sinceDays: parsedSinceDays.value ?? null })
     })
   ]
 })
@@ -0,0 +1,31 @@
+import { defineMcpClientConnection } from 'eve/connections'
+import { adminOnlyVercelAuth, VERCEL_PROJECT_ID, VERCEL_TEAM_ID } from '../lib/vercel-connect.js'
+
+const ALLOWED_TOOLS = [
+  'search_vercel_documentation',
+  'list_deployments',
+  'get_deployment',
+  'get_deployment_build_logs',
+  'get_runtime_logs',
+  'get_runtime_errors',
+  'get_project',
+  'list_agent_run_projects',
+  'list_agent_runs',
+  'get_web_analytics'
+] as const
+
+export const VERCEL_MCP_INSTRUCTIONS = VERCEL_TEAM_ID && VERCEL_PROJECT_ID
+  ? `**Vercel MCP connection (\`connection__vercel_mcp__*\`, admin/Slack/schedule only) — read-only, use judiciously:**
+- Discover exact schemas via \`connection__search\`, then call \`connection__vercel_mcp__<tool>\`.
+- The connection is pre-scoped to the \`nuxt-js\` team and the \`nuxt\` (nuxt.com website) project — \`teamId=${VERCEL_TEAM_ID}\`, \`projectId=${VERCEL_PROJECT_ID}\`. Pass both explicitly to \`list_deployments\`, \`get_deployment\`, \`get_deployment_build_logs\`, \`get_runtime_logs\`, \`get_runtime_errors\`, \`get_project\`, \`get_web_analytics\`.
+- Nuxi's own Agent Runs (\`list_agent_runs\`) use the same \`teamId\` but a DIFFERENT \`projectId\` — the \`eve\` service's own project, not the website. Call \`list_agent_run_projects\` first to discover it. Still NOT tokens/cost — use \`ai_gateway__*\` for that. No per-run trace access — this connection only exposes run-level metadata, never raw conversation content.
+- \`get_web_analytics\` (visitors/pageviews/custom events, production only): \`mode: 'count'\` (default) returns one total, e.g. "how many visitors this week"; \`mode: 'aggregate'\` groups by up to two \`by\` dimensions (hour/day/week/month/year, country, route, requestPath, referrerHostname, deviceType, browserName, eventName, flags/<name>, ...) and requires \`since\`+\`until\`. \`dataset: 'visits'\` (default) for pageviews, \`'events'\` for custom \`track()\` events. \`filter\` is OData, e.g. \`requestPath eq '/docs'\`. Requires Web Analytics enabled on the project.
+- \`search_vercel_documentation\` needs no ids — general Vercel platform docs search.`
+  : ''
+
+export default defineMcpClientConnection({
+  url: 'https://mcp.vercel.com/nuxt-js/nuxt',
+  description: 'Vercel platform for nuxt.com: deployments, runtime logs/errors, web analytics, and Nuxi\'s own Agent Runs observability. Admin/Slack/schedule sessions only.',
+  tools: { allow: ALLOWED_TOOLS },
+  auth: adminOnlyVercelAuth('Vercel MCP', { connector: 'vercel/mcp', principalType: 'app', autoProvision: false })
+})
@@ -1,13 +1,15 @@
 import { defineDynamic, defineInstructions } from 'eve/instructions'
 import { ADMIN_MCP_INSTRUCTIONS, canAccessAdminMcp } from './tools/admin-mcp.js'
+import { AI_GATEWAY_INSTRUCTIONS } from './tools/ai-gateway.js'
+import { VERCEL_MCP_INSTRUCTIONS } from './connections/vercel-mcp.js'
 import { buildInstructionsWithDate } from './lib/base-instructions.js'
 
 export default defineDynamic({
   events: {
     'session.started': async (_event, ctx) => {
       const auth = ctx.session.auth.current
       const markdown = canAccessAdminMcp(auth)
-        ? `${buildInstructionsWithDate()}\n\n${ADMIN_MCP_INSTRUCTIONS}`
+        ? [buildInstructionsWithDate(), ADMIN_MCP_INSTRUCTIONS, VERCEL_MCP_INSTRUCTIONS, AI_GATEWAY_INSTRUCTIONS].filter(Boolean).join('\n\n')
         : buildInstructionsWithDate()
 
       return defineInstructions({ markdown })

@@ -53,6 +53,8 @@ Do NOT call \`list-*\` first when the page is given — call the get tool direct
 - \`report_issue\` — call when you cannot resolve the user's question after exhausting all available tools, or when the user expresses frustration
 - ALWAYS respond with text after tool calls — never end with just tool calls
 
+**Restricted tools/connections:** Some connections (e.g. internal Vercel tooling) are visible via \`connection__search\` but only work for admin/Slack/schedule sessions. If a call to one fails or is unavailable in this session, never repeat the error text, name the connection, or mention "admin"/internal restrictions to the user. Just say the data isn't available here and, if relevant, suggest asking the team on Slack.
+
 **Web search:** Only use \`web_search\` when the user **explicitly** asks about recent events or real-time data beyond the Nuxt docs, or if \`search_github_issues\` returned no results. Never search proactively.
 
 **Web search queries:** Match the user's wording. **Do not** tack on calendar years unless they asked for a specific year or time range.

@@ -0,0 +1,22 @@
+import { connect, type EveAuthorizationOptions } from '@vercel/connect/eve'
+import type { SessionContext } from 'eve/context'
+import { canAccessAdminMcp } from './admin-mcp-access.js'
+
+export const VERCEL_TEAM_ID = process.env.NUXI_VERCEL_TEAM_ID
+export const VERCEL_PROJECT_ID = process.env.NUXI_VERCEL_PROJECT_ID
+
+export function adminOnlyVercelAuth(label: string, connectOptions: EveAuthorizationOptions) {
+  return (ctx: SessionContext) => {
+    if (!canAccessAdminMcp(ctx.session.auth.current)) {
+      return {
+        principalType: 'app' as const,
+        async getToken(): Promise<never> {
+          console.warn(`[vercel-connect] blocked non-admin access to ${label}`)
+          throw new Error('This tool is not available in the current session.')
+        }
+      }
+    }
+
+    return connect(connectOptions)
+  }
+}
@@ -0,0 +1,36 @@
+import { defineSchedule } from 'eve/schedules'
+import type { ScheduleHandlerArgs } from 'eve/schedules'
+import {
+  receiveOnSlack,
+  resolveSinceDays,
+  skillWorkflowMessage
+} from '../lib/workflows.js'
+
+const SKILL_ID = 'analytics-digest'
+const DEFAULT_WINDOW_DAYS = 7
+
+export async function runAnalyticsDigest({
+  receive,
+  appAuth,
+  sinceDays
+}: {
+  receive: ScheduleHandlerArgs['receive']
+  appAuth: ScheduleHandlerArgs['appAuth']
+  sinceDays?: number
+}) {
+  const windowDays = resolveSinceDays(sinceDays, DEFAULT_WINDOW_DAYS)
+
+  return receiveOnSlack({
+    receive,
+    appAuth,
+    message: skillWorkflowMessage(SKILL_ID, windowDays)
+  })
+}
+
+export default defineSchedule({
+  // 15 min after weekly-digest so both land close together for the Monday morning read.
+  cron: '15 5 * * 1',
+  async run({ receive, waitUntil, appAuth }) {
+    waitUntil(runAnalyticsDigest({ receive, appAuth }))
+  }
+})
@@ -33,7 +33,7 @@ export async function runFirehoseSummary({
 }
 
 export default defineSchedule({
-  cron: '0 9 * * 1-5',
+  cron: '0 5 * * 1-5',
   async run({ receive, waitUntil, appAuth }) {
     waitUntil(runFirehoseSummary({ receive, appAuth }))
   }

@@ -28,7 +28,7 @@ export async function runWeeklyDigest({
 }
 
 export default defineSchedule({
-  cron: '0 9 * * 1',
+  cron: '0 5 * * 1',
   async run({ receive, waitUntil, appAuth }) {
     waitUntil(runWeeklyDigest({ receive, appAuth }))
   }

@@ -0,0 +1,57 @@
+---
+description: Report on nuxt.com traffic trends, top pages/referrers, and doc pages that combine high traffic with poor feedback, for a recent window.
+---
+
+When producing the analytics digest (scheduled or on request):
+
+**Slack delivery:** Your reply IS the Slack message — Eve posts it verbatim to this channel. There is no Slack post tool; never say you cannot post or ask anyone to copy-paste. You are Nuxi (:nuxi:) giving the team a traffic pulse. Be concise, linked, and actionable — not a wall of numbers.
+
+- First line: **Nuxt traffic digest — last N days** plus date range in parentheses.
+- No preamble, no delivery disclaimers, and no meta wrap-up.
+- Use **bold** section labels — never markdown `#` headings.
+- ONE message only. Blank line between sections.
+- Every page/dashboard reference must include a clickable Slack link: `<https://…|label>`.
+- Sparingly use :nuxter: or :nuxt_cool: (0–2 total) when something is worth celebrating or urgent.
+
+**Data steps** (via `connection__vercel_mcp__get_web_analytics` — discover the exact schema with `connection__search` first if unsure. Pass `teamId`/`projectId` explicitly on every call. Counts default to production traffic only. "Previous window" = the N days right before the current window, same length, for deltas):
+
+1. `mode=count, dataset=visits` for the current window AND the previous window → total visitors/pageviews + delta %.
+2. `mode=aggregate, dataset=visits, by=['day']`, current window → daily trend, to spot spikes/drops.
+3. `mode=aggregate, dataset=visits, by=['route'], limit=10` for the current window AND the previous window → top pages with a per-page delta.
+4. `mode=aggregate, dataset=visits, by=['referrerHostname'], limit=8`, current window → top referrers.
+5. `mode=aggregate, dataset=visits, by=['country'], limit=5` and a separate call `by=['deviceType']`, current window → geo/device breakdown.
+6. `admin_mcp__feedback_stats` (`topPages=10`) and `admin_mcp__list_feedback` (`ratings=["not-helpful", "confusing"]`, `limit=20`) → candidate poorly-rated pages.
+7. For each candidate page from step 6: look up its traffic rank from step 3, or if it isn't in the top 10, run a targeted `get_web_analytics` `mode=count, filter="requestPath eq '<path>'"` → flag pages that are BOTH meaningfully-trafficked AND poorly rated.
+
+**Link cheat sheet:**
+
+- Docs page: `<https://nuxt.com/docs/…|Page title>`
+- Analytics dashboard: `<https://vercel.com/nuxt-js/nuxt/analytics|Vercel Web Analytics>`
+
+**Output template:**
+
+**Nuxt traffic digest — last 7 days** (Jul 15 – Jul 21, 2026)
+
+:bar_chart: **Traffic pulse**
+• *12,430 visitors* (+8% WoW), *31,200 pageviews*
+• Trend: steady, small bump on Jul 18 (release day)
+
+:page_facing_up: **Top pages**
+1. <https://nuxt.com/docs/getting-started/installation|Installation> — 2,100 visits (+12%)
+2. <https://nuxt.com/docs/guide/directory-structure/app|App directory> — 1,540 visits (-3%)
+3. …
+
+:compass: **Referrers & audience**
+• Top referrer: <https://google.com|Google> (48%), then <https://github.com|GitHub> (15%)
+• Top countries: US, DE, FR — mostly desktop (72%)
+
+:rotating_light: **Pages that need love** (high traffic + poor feedback)
+• <https://nuxt.com/docs/…|Rendering modes> — 1,800 visits this week, feedback 2.1/5 ("examples outdated")
+• If none qualify: one bullet ":large_green_circle: All clear — no high-traffic page flagged this week."
+
+Rules:
+- If a section has zero or flat data, say so in one bullet with a likely cause (low traffic, analytics not enabled, etc.) — do not skip the section.
+- Never list a page without its `<url|label>` link and its traffic number.
+- **Pages that need love** must cite the real feedback comment/score, not just a page name.
+- Never invent traffic numbers — if a `get_web_analytics` call fails or returns nothing, say so instead of guessing.
+- This is the deep traffic dive; `weekly-digest` only carries a one-line pulse — don't duplicate the full breakdown there.