cachly is now on awesome‑mcp‑servers — 20,000 downloads and what comes next

What awesome-mcp-servers is and why it matters

punkpeye/awesome‑mcp‑servers is a community-curated directory of production-ready MCP servers. Every listing is reviewed before merge — it is not an automated registry. Getting listed means your tool was evaluated for quality, usefulness, and reliability by maintainers who know the MCP ecosystem.

For discovery, the list is as important as npm. Developers looking for “MCP memory server” or “MCP persistent context” will find cachly there, alongside a small number of carefully selected tools. We are proud to be in that company.

The numbers — nearly 20,000 downloads

Six weeks ago we published our 12,000 downloads in 23 days post. The momentum has continued. As of today:

Growth has remained organic — no paid acquisition, no Product Hunt spike. The channel that works is MCP-native discovery: developers searching for “mcp server memory” on GitHub, npm, and lists like awesome‑mcp‑servers. With the listing merged, that surface area just got meaningfully larger.

What we shipped since the last post

The 12,000-downloads post captured the pivot to AI memory. In the six weeks since, we shipped four significant things.

1. Team Brain — shared memory for organisations

Individual brain memory is powerful. Team Brain is the multiplier. Any organisation can now create a shared brain that all members access through a single invite link. When one engineer fixes a tricky k8s race condition, every AI session in the team benefits — the fix is stored once, recalled everywhere.

Under the hood: team_learn writes to the org brain, recall_best_solution and brain_search query it. Role-based access control (owner / admin / member) keeps data private to the organisation. Free tier includes org brains up to the Free storage limit; Pro and above unlock larger shared stores.

2. CI-driven learning — the brain gets smarter on every deploy

We added a new endpoint — POST /v1/cache/:token/cls/ingest — that CI runners can call after builds, test runs, and deployments. A single GitHub Actions step after your test suite:

- name: cachly learn from CI
  uses: cachly-dev/cachly-action@v1
  with:
    token: ${{ secrets.CACHLY_CACHE_TOKEN }}
    event: ci_outcome
    status: ${{ job.status }}

Every CI run becomes a training signal. The brain stores which commits fixed which failures, which test patterns predict flaky deploys, and which architecture decisions have a track record. Future AI sessions can query this history with brain_predict_failures.

3. gRPC BrainService — AI-to-AI access

For teams building AI agents or orchestration systems, cachly now exposes a gRPC interface at grpc.cachly.dev:443. An AI agent can call BrainService.Recall, BrainService.Learn, and BrainService.Search directly over HTTP/2, without going through the MCP transport. This opens up multi-agent workflows where one agent teaches the brain and another recalls.

4. Failure prediction — prevent incidents before they happen

brain_predict_failuresanalyzes the causal knowledge graph accumulated from your CI history, past incidents, and IDE diagnostics. It surfaces patterns like: “every deploy that touched k8s/config.yaml without updating deployment.yamlcaused a namespace conflict within 24 hours.” The prediction runs locally against your brain — no external model calls.

Where to find cachly

cachly is now listed on every major MCP discovery surface:

• awesome‑mcp‑servers — the community standard for MCP quality
• Glama — remote server connector directory
• Smithery — install-in-one-click MCP registry
• npm — npx @cachly-dev/mcp-server@latest init

What comes next

The immediate roadmap has three items:

• Native IDE extensions — VS Code and JetBrains extensions that add Brain status to the status bar and offer one-click recall without switching to the MCP chat.
• Brain analytics dashboard— visualise what your team's AI has learned: most recalled patterns, top-confidence predictions, knowledge graph growth over time.
• Read-only org sharing — organisation members will be able to recall from the shared brain without being able to write. Useful for onboarding new engineers or external contractors.

The Discord username for the server-author flair request is pending — if you are the maintainer reading this: HeinrichNebula.

Try it in 30 seconds

The fastest path to a working brain: