☕ MCP Morning — Wed, Oct 14, 2026
Good morning. Today: a reverse-engineering toolkit that went from ~5k to ~270k weekly npm downloads, CAD you can drive from your agent, and a talk on making games legible to coding agents.
🔥 Trending & new MCPs
- REA — Reverse engineer anything from your terminal or agent, from app behavior down to native binaries, with one CLI and MCP server.
Why it matters: reverse engineering is a natural fit for agent loops (inspect, hypothesize, re-run), and this packages it as one tool your agent can call. Its npm package
rea-agentslogged 270,176 downloads last week, up from 5,324 the week before; it's also on the Hacker News front page today (98 points). - SolidCraft — An open-source parametric 3D CAD app written in pure Rust, for desktop and web, with automation through MCP. Why it matters: CAD has been one of the harder domains for agents to touch. A parametric model your agent can edit through MCP means "make the bracket 2 mm thicker" becomes a tool call. Show HN, 138 points.
- DBHub by Bytebase — A token-conscious database MCP server for Postgres, MySQL, SQL Server, Oracle, MariaDB and SQLite. Why it matters: one server for most SQL databases, designed to keep responses small, so schema lookups and queries eat less of your context window. 46,758 npm downloads last week.
- agent-device by Callstack — Mobile app automation and verification for AI coding agents: a CLI, an MCP server and a typed Node.js API for iOS, Android, HarmonyOS, TV, web, macOS and Linux. Why it matters: agents that write mobile code can now check their work on a real simulator or device instead of guessing. 345,287 npm downloads last week.
- jevmem — Saves the decisions, rules and failed approaches from your Claude Code chats to a
JEVMEM.mdfile in your repo, and brings the relevant ones back next session. Why it matters: project memory that lives in plain markdown in your repo, so you can read it, review it and commit it. New this month, 123 ⭐ on GitHub.
🗞️ Agent story
Your Game Is A Harness (HN) — Game dev Kyle Kukshtel argues that LLMs aren't bad at games, they just can't see them. His fixes: seeded randomness and save/load from day one, one "fat struct" object type, a command pattern so state changes can be replayed, game data outside the code so changes don't need a recompile, a simulation that runs headless, and text output an agent can read (odiff for screenshot diffs, samply for profiling). Worth reading even if you never ship a game: the same ideas make any codebase easier for an agent to work on.
🚀 Rising this week
Biggest week-over-week npm download growth in our rankings:
- REA — 270,176 downloads (+4,975%)
- Acquisition.gov — 10,650 (+2,418%)
- Concord MCP — 3,066 (+2,240%)
- tewip — 5,120 (+1,924%)
- outcometick — 5,999 (+1,614%)
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.