MCP Morning — Wed, Nov 11, 2026
Five MCP picks for Wednesday: the world's law as a git history your agent can query, a real radio receiver an agent can drive, an iPhone your agent can tap, a read-only door into Odoo, and GitHub's own server.
Trending & new MCPs
- Global law — Each country's law as its own git repository, where every commit is one law taking effect, gathered here as submodules. Its
tools/mcp_server.pyis an MCP server with 18 read-only tools: find a provision, read it as it stood on a given date, see which law wrote each word, compare dates and check a brief against the law in force. It runs locally on Python and git and, per the README, makes no network calls. Why it matters: "what did this clause say on that date" is exactly where models guess; here the answer comes from a dated commit. Tools under CC0 (law text under each country repo's terms), 1,930 GitHub stars. - rtl-sdr-mcp — Turns a cheap RTL-SDR USB dongle into agent tools: sweep the spectrum, look at waterfalls returned as images, demodulate and transcribe audio, watch a band for intermittent transmitters and decode ISM devices through rtl_433. The README walks through a session where Claude spends 10 tool calls trying to identify an unknown signal near 434 MHz and ends with an unproven hypothesis rather than a made-up answer. Why it matters: the model picks the frequencies and reads the plots itself instead of getting a recording after the fact. MIT, 6 stars, created Nov 10.
- phonectl-mcp — An MCP server on your Mac that lets an agent take screenshots, read on-screen text, tap, swipe, type and open apps on a real iPhone over USB, then Wi-Fi. Nothing is installed on the phone and no jailbreak is needed, but it requires iOS 27 or later with Developer Mode on. Why it matters: it covers apps that only exist on the phone. The README is blunt that an agent can do anything your finger can, including sending messages and paying, so keep instructions narrow. MIT, 3 stars, macOS only.
- odoo-mcp-lite — An Odoo module (17, 18 and 19) that exposes a read-only MCP endpoint at your ERP's own URL, with OAuth 2.1 built in, so Claude, ChatGPT, Claude Code or Cursor can search, count and total records as the signed-in user. It never returns passwords, tokens or binary fields, and has a per-user rate limit plus a switch to pause all AI access. Why it matters: read-only and permission-bound is the sane first step for letting an assistant near accounting data. Apache-2.0, 2 stars, not affiliated with Odoo S.A.
- GitHub MCP Server — GitHub's official server for reading code, managing issues and pull requests, following Actions runs and reviewing security findings, available as a remote server hosted by GitHub or as a local build. Version 2.1.0 shipped Nov 3. Why it matters: it is the default way most coding agents reach GitHub, so its toolsets and permissions are worth a look before you connect it. MIT, 34,386 stars.
Agent story
Modern Web Guidance: get your coding agent off ~2019 JavaScript (HN) — The Chrome team, with the Edge, Firefox and MDN teams, published a set of agent skills and a CLI that feed curated guidance on current web platform features into a coding agent's context. The pitch is that models default to older patterns because their training data is full of legacy code, so they write heavy JavaScript and polyfills for things browsers now do natively. You can search a guide, retrieve one by ID, or run legacy-scan over a codebase to find patterns to modernize, and the guides are pruned with evals to stay token-efficient. 368 points on Hacker News.
Rising this week
Biggest week-over-week download growth in our rankings:
- anytype-mcp — 18,143 downloads (+1043%)
- mcp-dnstwist — 840 (+637%)
- Three-Live 3D, CAD & Spatial XR Compiler — 2,050 (+466%)
- slg — 1,671 (+466%)
- Cargo — 80,935 (+457%)
directory.json was generated 2026-11-11; npm weekly counts run through Nov 9. 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.