← Blog
Milestone · May 2026

We're the #1 MCP memory server on npm

472 weekly installs. More than the next four memory MCP servers combined. Here's what worked, what didn't, and the lesson from our first real signups.

The numbers (May 23, 2026)

npm packageWeekly installs
@cachly-dev/mcp-server472
mcp-memory-ts27
memory-mcp-lite20
codex-mcp-memory-server19
mag-memory17

Source: npm registry downloads API, May 16–22, 2026.

That's 5.7× the runner-up, and more than the next four memory-focused MCP servers combined. The total last-30-days window is 11,161 downloads.

The first real users showed up this week

We hit a stranger milestone this week than the npm one. A real human we don't know found cachly, signed up, and used it. Then a second one subscribed to the newsletter from a different country. No launch post, no Show HN, no Product Hunt. They each got there through search and MCP discovery.

We've been writing about cachly for two months. This week was the first time we got to watch our own product receive the kind of quiet, unprompted signups that mean: this solves a real problem for someone.

What we think drove the lead

1. We picked one job and did it ridiculously well

The other memory MCPs are general key-value stores or vector databases with an MCP wrapper. cachly does one job: make your AI act like it remembers yesterday. Every tool, every default, every doc page is pointed at that outcome.

That focus shows up in concrete features: causal_trace walks the Causal Knowledge Graph to find the actual root cause of a bug. memory_crystalize distills 200 lessons into a briefing your AI can absorb in one tool call. brain_predict warns the AI before a known failure pattern is about to repeat. None of those exist in the other memory MCPs.

2. Zero-friction onboarding

The single biggest growth lever: npx @cachly-dev/init detects every installed editor (Claude Code, Cursor, Windsurf, Cline, VS Code, Copilot, Zed) and writes the MCP config for each of them automatically. No copy-paste. No JSON editing. The OAuth device-flow login takes one browser click — no password, no API token to copy.

We measured drop-off at every step before MCP discovery. The 30-day conversion from install to first lesson stored went from 14% (when users had to edit JSON manually) to 71% (after we shipped the init wizard).

3. Ambient learning — the AI never asks the user to remember

Every other memory system has the same flaw: it asks the user (or the AI) to explicitly call save_memory(). That never works in practice. AIs don't remember to remember.

cachly turned this inside out. We install a git post-commit hook during setup. Every commit auto-extracts a lesson. Every session_end fires automatically when the editor closes. Every session_start automatically loads relevant past lessons before the AI even responds to the user.

The result: 1,247 lessons stored per month per active brain on average. The next-closest memory MCP, when we tested with the same user, stored 0 — because the AI never called the save tool.

4. We shipped on the right calendar

2026 is the year “context engineering” replaced “prompt engineering” as the headline skill. Every major editor — Claude Code, Cursor, Windsurf, Zed — added MCP support in the last six months. Anthropic published the official MCP registry in March. Developers started searching for “mcp memory” and “persistent context” at 4× the rate of January.

We didn't cause the wave. We were just early enough to put a really good package at the top of the npm search results when the wave hit.

What didn't work

What's next

We shipped @cachly-dev/[email protected] alongside this post: 84 verified tools, 350 passing tests, a tightened description for npm search, and 20 new keywords aimed at the 2026-era search vocabulary (context-engineering, agent-memory, long-term-memory, rag-memory).

We're also opening up the Knowledge Commons— verified lessons that any cachly user can publish (with provenance) and any other cachly user can import. The first cross-team brain transfer happened last Thursday. We'll write that one up separately.

cachly is a persistent AI Brain for developers — memory shared across Claude Code, Cursor, GitHub Copilot & Windsurf simultaneously. Auto-detects every editor. Bootstraps from your git history. 89 MCP tools. Free tier, EU servers, no credit card.

Your AI is forgetting everything right now.

Every session starts blank. Every bug re-discovered. Every deploy procedure re-explained. cachly fixes that in 30 seconds — your AI remembers every lesson, every fix, every teammate's hard-won knowledge. Forever.

🇪🇺 EU servers · GDPR-compliant🆓 Free tier — forever, no credit card⚡ 30-second setup via npx🔌 Claude Code · Cursor · Copilot · Windsurf

Try it

One command. No credit card. No expiry.

npx @cachly-dev/init

Or read the full docs first.