Reduce MCP Find load and preserve useful content

[gusmar2017]

Problem and result

MCP Find could turn transient database failures into cached empty pages, scan more data than needed, rewrite unchanged records during sync, and expose broad bot-query surfaces. This change makes outages retryable, cuts recurring database and cache work, preserves the current 961 useful server pages in search, and improves setup guidance and measurement.

What changed

• Return uncached 503 responses for database failures while preserving real 404s.
• Add bounded database deadlines, query inputs, cache keys, directory pagination, and middleware memory.
• Move directory pages and filtered search onto isolated ISR routes; cache sitemap and LLM discovery outputs.
• Require meaningful README or tool evidence for index eligibility and prefilter cold scans before exact evaluation.
• Avoid sync writes when source-owned fields are unchanged; add retry, pagination, partial-failure, and cache-invalidation safeguards.
• Add source-derived client configuration guidance and emit analytics only after successful clipboard copies.
• Correct homepage and badge claims, track blog-to-server clicks, improve related articles, and refresh five sourced guides.
• Strengthen the content release gate and retain the existing cache-invalidation guard improvements.
• Notify IndexNow about canonical server URLs changed by the daily sync and blog URLs changed on main; publish the required verification key and batch submissions within the protocol limit.

No database migration is required. The code continues to support installations without the optional README-length columns.

Validation

• Full credentialed production build: passed, 1,165 pages and 961 server details
• Type checks: passed, 5/5 tasks
• Web tests: passed, 29 files and 382 tests
• Lint: passed with one pre-existing hook dependency warning
• Fork-safe compile-only CI path: passed
• Blog release and MDX/link/JSON checks: passed
• Cache invalidation guard: passed
• git diff --check and secret scan: passed

Operations note

The MCP Find Supabase project is now on Small compute. Early post-upgrade metrics show materially more available memory and essentially no swap use; retain or downgrade only after the scheduled observation period and representative traffic.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
mcp-find-web	Ready	Preview	Sep 10, 2026 9:52pm UTC