MCP Morning — Sun, Oct 18, 2026

Good morning. A Sunday mix: Sentry's iOS build server under a new name, a keyless local web layer for agents, a desktop SQL workspace with its own MCP server, a reference checker for papers and LAN-only control of Tapo smart-home gear. Plus a map of how little real work AI benchmarks cover.

Trending & new MCPs

• MobileBuildMCP by Sentry — The project formerly known as XcodeBuildMCP (the old repo URLs now redirect here): an MCP server and CLI that let agents build, run and test iOS and macOS projects with simulators and devices. Why it matters: one package gives you both modes, installable with Homebrew or npx -y mobilebuildmcp@latest mcp, and mobilebuildmcp init adds optional agent skills for the MCP tools or the CLI. It needs macOS 14.5+ and Xcode 16+, device tools need code signing set up in Xcode, and it sends runtime error telemetry to Sentry unless you opt out. 6,488 ⭐ on GitHub.
• wigolo — A local-first web layer for agents: search, fetch, crawl, extract, cache, find-similar and research loops over MCP or REST, with no API keys for the core tools. Why it matters: search fans out across 18 direct engine adapters with rank fusion and an on-device reranker, every response is cached under ~/.wigolo/, and failures are labeled in the result (a bot-protected page comes back as blocked_by_challenge, not as content). npx wigolo init --agents=claude-code,cursor sets it up and wires your agents; it needs Node 20+ and about 1.5 GB of disk. Public beta, AGPL-3.0. 5,482 ⭐ on GitHub.
• Tabularis — An open-source desktop SQL workspace (Rust and Tauri) with 3 built-in drivers and 24 shipped plugins, including DuckDB, ClickHouse, Redis and Firestore, and a built-in MCP server. Why it matters: tabularis --mcp exposes your saved connections, so an agent can list databases and tables, describe schemas and run queries from the same app you browse data in; Settings has one-click config for Claude Desktop, Cursor and Windsurf. Results can come back as JSON or the more compact TOON format. Note that run_query executes any SQL, so point it at connections you are comfortable letting an agent write to. 5,230 ⭐ on GitHub.
• paper-preflight — Checks every reference in a paper (LaTeX, Word, Markdown, Typst, PDF or a plain list) against Crossref, dblp, arXiv, DataCite, PubMed and OpenAlex, with an optional MCP server for coding agents. Why it matters: it flags references that don't exist, don't match what you wrote, were retracted, or are preprints that have since been published, and no language model takes part in a verdict; when it can't tell, it says "cannot determine". The project claims 0.79 false alarms per 100 references on held-out arXiv papers. uvx paper-preflight check path/to/paper runs it without installing. New this month, 946 downloads in the last seven days.
• tolpo — Local-first control of TP-Link Tapo devices for agents: lights, plugs, hubs and their sensors, cameras, doorbells and more, as an MCP server and an HTTP API, talking to each device over your LAN through python-kasa. Why it matters: every action has a level from 0 to 4. Reading and reversible control (power, brightness, color) are on by default; settings and security changes stay off until you raise the level, and pairing or a factory reset also need a confirmation phrase on each call. Wi-Fi names, passwords and tokens are never returned. New this week, 1,078 downloads in the last seven days.

Agent story

A benchmark map of the US economy (HN) — Mercor mapped 29 agentic benchmarks onto the US Department of Labor's O*NET taxonomy of 923 occupations and 18,797 tasks, counting a task as covered when an LLM judge (checked against human expert tags) maps at least 5 benchmark prompts to it. Only 924 tasks and 208 occupations clear that bar, and Mercor estimates that 90% of American working hours go to tasks with little or no benchmark coverage. Coverage is not random either: it skews toward the largest, most AI-exposed labor markets, so lawyers' tasks are heavily covered while legal secretaries and court clerks barely appear. For anyone building agents for ordinary office work, it is a reminder that leaderboard scores say little about most of the jobs you might be automating. 610 points on Hacker News.

Rising this week

Biggest week-over-week npm download growth in our rankings, for the week ending Oct 16:

1. Anna's Archive — 2,058 downloads (+914%)
2. cachly-mcp — 5,595 (+707%)
3. DeXe DAO governance — 731 (+596%)
4. figma-ui-mcp — 917 (+443%)
5. aicommander — 975 (+436%)

Downloads include CI and bot traffic, so treat these as signals, not endorsements. How we rank →

MCP Morning lists servers from public data (the official MCP Registry, GitHub, npm and PyPI). We don't run or test them, so review the code and permissions before you connect one to your agent.

Get it every morning