# ☕ MCP Morning — Wed, Oct 28, 2026

Good morning. Today's picks: a 3D building editor that agents can drive, analytics for people who run MCP servers, a way to hand legwork to cheaper coding agents, a hosted e-shop scraper, and a shared memory for every agent on your PC. The story is a developer's take on what you still need to know when agents write the code.

## 🔥 Trending & new MCPs

- **[Pascal Editor](https://github.com/pascalorg/editor)** — An open-source, local-first 3D building editor built with React Three Fiber and WebGPU that runs in the browser or from the CLI, with MCP so AI agents can work on the model.
  *Why it matters:* `npx @pascal-app/cli editor` starts the editor plus an authenticated MCP service on loopback ports and keeps projects in a local SQLite file; agents connect with `pascal mcp connect`. The README suggests one agent client per local service, since clients share the active scene. There's also a self-hosted container. MIT, 25,161 ⭐ on GitHub.
- **[mcpspan](https://github.com/mcpspan/mcpspan)** — Open-source, self-hosted analytics for MCP servers: which tools, resources and prompts get used, by which client, how fast they answer and why they fail.
  *Why it matters:* besides latency and errors, it tracks things that are specific to agents, like how many tokens your tool list costs each session, tools agents are shown but never call, tools they ask for that you don't have, and repeated identical calls that point to a loop. Its own MCP server (`mcpspan-mcp`, read-only) lets you ask your agent about those numbers, and there's a public demo to try it. MIT, 920 npm downloads in its first week.
- **[Pitroom](https://github.com/atastech/pitroom)** — Lets your main coding agent hand reading, fixes, tests and reviews to cheaper worker CLIs running in parallel, and get back only the answer, the diff and a receipt.
  *Why it matters:* workers can be OpenCode, Codex CLI, Claude Code, Gemini CLI, Qwen Code, Kilo Code or Antigravity, using whatever model that CLI defaults to, with a fallback chain when a free model hits rate limits. Every `path:line` a worker cites is checked on disk, and the project says workers can't commit, push, reset or bulk delete. MIT, 555 npm downloads last week.
- **[DeepScout MCP](https://github.com/deepscout-ai/deepscout-mcp)** — A hosted server that scrapes product catalogs from any e-shop and matches the same products across shops without needing an EAN or GTIN.
  *Why it matters:* it's a remote Streamable HTTP server with OAuth sign-in, so there's nothing to install. It's a paid service: only `scout_approve_job` spends credit, and the README asks you to keep your client's confirmation prompt on for that tool. Created two days ago; 6 ⭐ on GitHub.
- **[Turnbook](https://github.com/goyohan0611-png/graph-mind)** (listed as Graph-MIND) — Long-term memory that every coding agent on your PC reads and writes, stored word for word locally, with no LLM call to save anything.
  *Why it matters:* decide something with Claude Code and ask Codex about it later; each memory records which app and session it came from. The project claims 88.8% on LongMemEval and says secrets are masked before storage, with no account or telemetry. Note the one-line installers pipe a remote script into your shell, so read it first. Apache-2.0, 6,014 PyPI downloads last week.

## 🗞️ Agent story

**[Here Be Monsters: What Do You Still Need to Know When Agents Do the Work?](https://blog.adrq.dev/here-be-monsters-what-do-you-still-need-to-know-when-agents-do-the-work)** ([HN](https://news.ycombinator.com/item?id=50258868)) — Adrian Quiroga starts with a CI failure: SonarQube flagged an `App.tsx` where Codex had written whole components on single lines. The logic turned out to be correct, but it was code he had never read. His argument is that owning a system will depend less on having written it and more on knowing enough to notice when something is off, and he names three things to keep: fundamentals (so you recognize, say, quadratic slowdowns), breadth (so you know where to look) and "just-in-time depth", the willingness to dive into unfamiliar code when it matters. His example is fixing lifecycle bugs in his own agent orchestrator after remembering "level-triggered state machine" from a Kubernetes book. 173 points on Hacker News.

## 🚀 Rising this week

Biggest week-over-week download growth in [our rankings](https://mcpmorning.com/mcp/rising/), skipping one server we covered in the last week:

1. **[Agent Comms](https://mcpmorning.com/mcp/exadev-agent-comms/)** — 4,067 downloads (+1818%)
2. **[wigolo](https://mcpmorning.com/mcp/knockoutez-wigolo/)** — 2,831 (+828%)
3. **[WebPerf Snippets](https://mcpmorning.com/mcp/nucliweb-webperf-snippets/)** — 1,330 (+787%)
4. **[caldav-mcp](https://mcpmorning.com/mcp/dominik1001-caldav-mcp/)** — 2,029 (+782%)
5. **[iMessage History](https://mcpmorning.com/mcp/anipotts-imessage-mcp/)** — 836 (+696%)

npm counts now run through Oct 26. Downloads include CI and bot traffic, so treat these as signals, not endorsements. [How we rank →](https://mcpmorning.com/methodology/)

---

*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.*