Beat: publicly available, usable-now developer artifacts — tools/skills/MCPs we can ADOPT instead of build. Not news (ai-watch), not model releases (model-watch), not research (Sol). Run by code-eth, weekly. Reads first: watchlist.md · Apps/app-architecture_roadmap.md · eth-memory/shared/intelligence-board.md.

Honest framing: watchlist is empty this week (both 06-12 entries resolved 06-19), so this was a generic-sweep week. The aggregator listicles were the usual noise ("345 skills! 14,000 servers!"), but the needs-specific sweep landed one genuinely strong ADOPT that retires a roadmap Priority-2 build, plus a material update to a long-pending carry (OpenClaw cooperation). One generic ADOPT-CANDIDATE boarded, well under the 3/week cap. Finance finds were real but out of Code's lane (no portfolio/finance — commit 6bdf5c6) → flagged for Sol, not boarded.

WATCHLIST

Open: none. All entries resolved as of 06-19 (Obsidian MCP → ADOPT; ClawCode → SKIP). Nothing new dropped by Zaina or the voices this week. Watchlist Resolved section unchanged.

ADOPT-CANDIDATE (generic — 1 this week)

ntfy + ntfy-mcp-server (cyanheads) → the off-the-shelf Communication Hub we were about to hand-build

• MCP repo: https://github.com/cyanheads/ntfy-mcp-server · Apache-2.0 · ⭐16 (thin wrapper — low stars, but see below) · v2.1.1, 2026-06-12 (71 commits, actively maintained, Vitest + CI, built on @cyanheads/mcp-ts-core). Same author as the Obsidian Node alternative noted last week — known-good MCP builder.
• Underlying service: ntfy.sh (https://ntfy.sh, self-hostable, ~20–50 MB RAM, has its own iOS/Android app) — the well-established HTTP pub/sub push service. POST to a topic → every subscribed phone gets it instantly. Self-host with Docker; protected topics use bearer token / basic auth.
• WHAT: ntfy is a tiny self-hosted push-notification server with a phone app. The MCP server gives the voices 4 tools — ntfy_publish_message (send/update, full param coverage), ntfy_manage_message (clear/delete by ID), ntfy_fetch_messages (poll with filters), ntfy_search_emoji_tags — plus an ntfy://{topic} resource. Self-hosted support is first-class: NTFY_BASE_URL / NTFY_SERVERS env vars with per-server scoped auth.
• WHY US: This is the highest-value kind of find — it replaces a planned build. Roadmap Priority 2 "Communication Hub" = "Push notifications to phone (replaces Telegram bot)... needs service worker + push API." ntfy deletes that whole build: there's already a maintained phone app and a one-line POST to a topic — no service-worker/VAPID/push-API plumbing for us to write and own. And it's strictly better than the current @eth_care_circuit_bot Telegram path: self-hosted = no third-party bot to depend on (local-first-push), and the MCP layer means any voice can push to Zaina's phone as a native tool call — the mesh monitors (sync-monitor / task-digest / gentleman-loop / comment-tracker, mesh-agents-monitoring) and the gentleman-loop check-ins could surface to her phone without the Telegram bot in the loop. Passes the cooperation test by construction: it's a channel the voices call, not a gateway that owns the loop.
• Caveats: (1) MCP wrapper is young/thin (⭐16) — but the value is ntfy itself (mature, ~20k★ class project) + a small bridge; if the bridge ever rots, a hand-rolled POST is trivial. (2) Self-host the ntfy server — don't rely on public ntfy.sh for anything private (topics are world-readable unless auth'd). (3) publish/manage are write tools — scope topics and auth before wiring to always-on monitors.
• Verdict: ADOPT-CANDIDATE. Spike: docker run ntfy on the Pi (or a Cloudflare-fronted box), install the ntfy phone app, point ntfy-mcp at it with a private auth'd topic, have one mesh monitor push a test ping to Zaina's phone. Cheapest possible proof of the Communication Hub before we build a service worker we don't need. Adopter: Eth-Code (infra) → Eth-Pi (host).

WATCH (real, relevant, not yet a decision)

• freema/openclaw-mcp — the OpenClaw cooperation question now has an answer path. https://github.com/freema/openclaw-mcp — "secure bridge between Claude.ai and your self-hosted OpenClaw assistant with OAuth2 auth." The pending spike since 06-12 was "does OpenClaw cooperate with our MCP servers or want to own the loop?" — a dedicated MCP bridge now exists, which means OpenClaw can sit behind an MCP boundary (Claude calls IN to it over OAuth2) rather than fronting our voices. OpenClaw itself shipped 2026.6.11 (now ~380k★, Slack relay, per-DM model overrides, file-driven --message-file). This materially de-risks the multi-channel-chat-gateway slice but doesn't retire the spike — still need to confirm it cooperates with eth-state specifically (no public eth-state mention, expected). Note: ntfy (above) already covers the push-to-phone slice MCP-natively and more simply, which keeps OpenClaw's urgency low. WATCH — pull forward only if/when we want the full Telegram/Signal/WhatsApp multi-channel gateway, not just push.
• Trend-intelligence MCPs (trend-pulse, TrendsMCP, Trends-MCP) — claude-world/trend-pulse (20–37 sources, MCP + CLI + Python, zero-auth free tier), trendsmcp/trends-agent-claude (Google/YouTube/TikTok/Reddit/Amazon/App Store), rugvedp/Trends-MCP (incl. Instagram Reels). These touch the same gap as the still-pending last30days skill — the social/community/IG-adjacent signal layer that is literally Zaina's "my Instagram is full of tools we haven't thought of." WATCH, paired with the last30days debt below — if we adopt a social-signal collector for this beat, compare these against last30days rather than running a hand-rolled web sweep forever.
• Agent-ops connectors entering Anthropic's directory (AgentGuard = guardrails that block malicious skills/data-leaks; Origin = fleet visibility into what agents do; Mnemoverse = cross-tool persistent memory; AgentMail = inboxes agents own; Alpic = MCP hosting). The directory is shifting from "connectors for Claude" → "infra for running agents in production." AgentGuard is the only one adjacent to a real need (we keep pulling in third-party skills; it overlaps Bumblebee's read-only-scan niche from edition 1) — WATCH as a possible complement to Bumblebee, not a replacement. The rest (memory, email, hosting) we either already own or don't need.

OUT OF LANE — flagged, NOT boarded (finance/portfolio = Sol)

• Personal Finance Engine (gregoryalan, 20 MCP tools, imports brokerage CSVs, net-worth/allocation/projections, local SQLite, no external APIs) and Finlynq (AGPL-3.0, self-host Docker, first-party MCP) both map cleanly to roadmap Priority-2 "Portfolio Dashboard (needs IB data pipeline)." Code stays out of portfolio/finance by the 06-26 lane decision (commit 6bdf5c6) — handed to Sol as a research pointer, deliberately not an ADOPT on this beat. Alpaca's MCP supports live trade execution — noted only to be explicit we would never wire that to an autonomous voice.

PENDING FROM LAST WEEK (honest carry-forward)

• last30days trial — STILL NOT DONE (now slipped 3 editions: 06-12, 06-19, 06-26). This sweep was again hand-rolled web search. The cost is exactly visible this week: the trend-intelligence MCPs above and the IG-signal gap are precisely what a proper social-layer collector would have surfaced first instead of me finding them sideways. Action stands and is now the top process debt: install last30days (or one of the trend-pulse-class MCPs) as a shared skill and run it before next week's sweep.
• OpenClaw cooperation spike — PARTIALLY ADVANCED. freema/openclaw-mcp (above) gives it a concrete path; the half-day eth-state-cooperation test still hasn't been run, but it's now lower-stakes since ntfy covers the urgent push slice.

Bottom line

Empty watchlist, so a generic week — but not a thin one. ntfy + ntfy-mcp is a real ADOPT: a self-hosted, MCP-native push service that retires the roadmap's Communication Hub build entirely (no service worker to write), beats the current Telegram bot on the local-first axis, and lets any voice push to Zaina's phone as a tool call — the mesh monitors and gentleman-loop check-ins get a phone surface for free. The long-pending OpenClaw question gained an answer path (openclaw-mcp, OAuth2 bridge) but stays WATCH since ntfy covers the urgent slice more simply. Finance/portfolio finds were good but handed to Sol, not boarded (lane discipline). Two process debts remain, and last30days has now slipped three weeks — it's the single highest-leverage upgrade for this beat and it keeps not happening.

Next run: install last30days (or a trend-pulse-class social collector) and run it BEFORE the sweep — three editions of "still hand-rolled" is enough.