MCPVault

We launched MCPVault, an independent directory for MCP servers, on Fazier today and it won Product of the Day.

The best part was not the badge. The launch comments gave us a very clear product roadmap, so we treated them like issues and shipped fixes during the launch window.

What we added today:

• Last verified date on every verified listing, based on the last successful live test.

• Live test evidence: MCP protocol revision, server version, tool count, and verifier runtime.

• Weekly re-verification: a verified server must still complete initialize plus tools/list. If it fails, the badge lapses automatically.

• Maintainer activity filters: browse servers updated in the last 30, 90, or 365 days.

• Auth method filters: owners can declare whether a server needs no auth, an API key, OAuth, or another method.

• Clearer grade disclosure: unclaimed listings now explicitly say their grade is automated from public GitHub signals, not maintainer-reviewed.

• Trust cadence copy: grades and repository signals refresh nightly, live tests run weekly.

• A production bug fix: our nightly GitHub refresh was overwriting owner-edited listing details. Fixed before it could cause damage.

The broader idea behind MCPVault is that a directory should distinguish between “this repository exists” and “this server actually responded to a real MCP call recently.”

Still early, but this was a good reminder: launch feedback is most valuable when you can convert it into shipped product improvements quickly.

Background: I run a small portfolio of SaaS products (VNDLY for inventory, Lapsewise for renewals, directree, a software directory). When Smithery, the leading independent MCP registry, got acquired in August 2026 and pivoted enterprise, I saw a gap: the MCP ecosystem (tens of thousands of servers, adopted by Anthropic, OpenAI, Google, Microsoft) had no independent quality directory left.

So I built MCPVault (mcpvault.io).

What I have shipped:

- Auto-ingestion from GitHub: 5,000+ MCP servers live, refreshed nightly.

- Quality grades A to F (recency, adoption, license, docs signals).

- README enrichment: install commands, client compatibility, tool extraction.

- 26,000+ tool-level SEO pages (every tool inside a server gets its own page).

- GitHub OAuth claim flow: maintainers prove push access and take control of their listing.

- Owner listing editor: description, category, topics, compatibility, install command, homepage, logo upload.

- A real verification harness: spawns the server sandboxed, requires an actual MCP stdio handshake (initialize + tools/list), re-checks weekly, auto-lapses on failure.

- Status-aware embeddable badges (Listed / Claimed / Verified) that update everywhere when status changes.

- Do-follow website links for verified servers.

The model: listings auto-ingest as minimal "stale" pages with a claim CTA. Claiming is free. Verification is free during early access (paid tier planned later, but the badge stays earned-only: you cannot buy it, your server has to pass the handshake).

The moat: nobody else actually runs the servers. Existing directories do static GitHub analysis at best. "Verified & Working, re-checked weekly" is a different product.

Happy to talk about the ingestion pipeline, the sandboxed verification harness, or the claim-funnel model.