MCP Morning — Fri, Oct 16, 2026
Good morning. Five servers that put scraping, IDE-grade edits, diagrams, a local code graph and a React Native health check in reach of your agent — plus Liquid AI's new MoE built for on-device tool use.
Trending & new MCPs
- Apify MCP Server — Lets an agent call any Actor on Apify Store (scrapers for social, maps, search, e-commerce and more) as an MCP tool, via the hosted endpoint at mcp.apify.com or local stdio. Why it matters: the server discovers Actors, loads their input schemas and returns run results, so agents can scrape or fetch without writing crawlers. Hosted mode supports OAuth; agentic payments (AGI, x402, Skyfire) can cover runs without a long-lived API token. 11,497 ⭐ on GitHub.
- Serena — An MCP toolkit that gives coding agents IDE-like semantic retrieval, editing, refactoring and (with the JetBrains plugin) debugging, at the symbol level rather than by line numbers. Why it matters: agents rename, find references and replace symbol bodies in one call instead of multi-step text surgery. It supports 40+ languages via language servers, or JetBrains' own analysis through a paid plugin. 30,310 ⭐ on GitHub.
- Next AI Draw.io — Draw and edit real draw.io diagrams by chat: web app, desktop app, and an MCP server (
@next-ai-drawio/mcp-server) for Claude Code, Cursor, Codex and VS Code. Why it matters: agents produce architecture and sequence diagrams on a live draw.io canvas (AWS/Azure/GCP/K8s icons included), with version history, selection-aware edits and export to.drawio, PNG or SVG. 36,273 ⭐ on GitHub. - Graphicle — A single Rust program that builds a local code graph and project memory (docs, decisions, past sessions) and serves both over MCP, with cited excerpts inside a token budget you set.
Why it matters:
graphicle setupindexes the repo and wires Claude Code, Cursor, Codex and others; no API key, account or telemetry for the graph and recall path. Show HN yesterday; new this month, 180 ⭐. - NativeKeel — Health check for React Native and Expo apps: upgrade blockers, store rules, crash risks, security and accessibility, as a CLI and as an MCP server agents can query before they touch native deps.
Why it matters: first published to npm this week (listed in the official MCP Registry). Agents get
scan_project,upgrade_plan,explain_errorand related tools instead of guessing New Architecture or Play/targetSdktraps. 9,942 npm downloads in the last seven days.
Agent story
LFM2.5-24B-A2B: Built for fast on-device agents (HN) — Liquid AI released LFM2.5-24B-A2B, a 24B-parameter MoE with 2B active per token, aimed at private agents that reason and call tools on-device. The post says it leads or matches larger peers on five of eight agentic/tool-use benchmarks (including 48.09 on LiveClaw vs 38.19 for Qwen3.6-35B-A3B, and 27.69 on AutomationBench vs 8.72), decodes at 184 tokens/s on an Apple M5 Max, and ships open-weight with llama.cpp, MLX and vLLM support. Knowledge and coding benchmarks still favor bigger models; Liquid recommends pairing it with retrieval or search. 285 points on Hacker News.
Rising this week
Biggest week-over-week npm download growth in our rankings:
- OMEGA Memory — 2,534 downloads (+1,408%)
- Summer Engine — 6,574 (+886%)
- REA — 270,176 (+4,975%)
- token-optimizer-mcp — 5,764 (+396%)
- Acquisition.gov — 10,650 (+2,418%)
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.