TL;DR
The AI Agent Builder Stack is 6 MCPs (Tavily, Perplexity, Firecrawl, Supabase, Redis, GitHub) that cover search, scraping, memory, and caching for any production AI agent. One install replaces a DIY infra stack that would take 2 weeks to build. Free to mid-scale; you only pay for API calls at the underlying services.
The AI Agent Builder Stack
Search, scrape, memory, and cache — the infrastructure every agent needs
Install the whole stack
$ mcpizy install tavily perplexity firecrawl supabase redis githubOne command installs and configures all 6 MCPs for Claude Code, Cursor, Windsurf, or any MCP-compatible client.
Why this stack?
Every production AI agent needs four primitives: real-time web data (Tavily + Perplexity), deep content extraction (Firecrawl), persistent memory (Supabase + pgvector), and fast caching (Redis). Without these, you're rebuilding the wheel — with them, your agent is production-grade on day one.
Why this specific combo: Tavily is fast and cheap for breadth, Perplexity is the gold standard for depth with citations, Firecrawl is the best open-source scraper, Supabase's pgvector gives you a free vector DB up to 500MB, and Redis is the standard for tool-level caching. Six MCPs replace a stack that would otherwise cost $500–2000/mo in SaaS (Pinecone + Serper + Apify + Upstash).
MCPs in this stack (6)
What this stack lets you do
Research agent that monitors 50 blogs weekly
See recipe- 1Firecrawl crawls each blog on a cron
- 2New posts embedded via Supabase pgvector
- 3Tavily queries across embedded corpus semantically
- 4Redis caches last 100 queries (24h TTL)
- 5Agent summarises weekly digest, posts to Slack
Deep-research agent with citations
See recipe- 1User asks a complex question
- 2Perplexity returns pre-summarised answer with citations
- 3Firecrawl pulls full text of top 3 cited sources
- 4Supabase stores the research for future reference
- 5Agent synthesises and returns with linked sources
Coding agent that reads and ships
- 1User asks 'fix bug in /api/checkout'
- 2GitHub MCP reads the file + recent commits
- 3Supabase pgvector finds similar past fixes
- 4Agent proposes + writes patch
- 5GitHub MCP opens PR, Redis caches the bug fingerprint
Estimated value
Replaces ~$800/mo of infra SaaS (Pinecone starter, Apify, Serper, Upstash) for a mid-scale agent. Cold-start savings: ~2 weeks of engineering time not spent on custom infra.
Frequently asked questions
Do I really need both Tavily and Perplexity?
For most agents, yes. Tavily is cheap and fast for broad search (30 results in 2s, $5/1000 queries). Perplexity is premium — pre-summarised answers with citations (better for final-answer quality). Use Tavily for ideation/breadth, Perplexity for commitment/depth.
Why Supabase over Pinecone for vector storage?
Cost. Supabase (pgvector) is free up to 500MB and ~$25/mo up to 8GB. Pinecone starts at $70/mo for comparable capacity. Unless you need <10ms p99 at 100M+ vectors, Supabase wins on price.
Can I skip Redis if I'm already using Supabase?
Short answer: yes. Long answer: Redis is much cheaper for ephemeral cache (TTL-bounded, no schema). Use Supabase for durable state, Redis for 'I called Tavily 5 min ago, use that'. The two-tier pattern saves ~40% on API bills.
Is this stack suitable for multi-agent systems?
Yes — each agent gets a subset of these MCPs. A 'researcher' agent gets Tavily+Firecrawl+Supabase; a 'writer' agent gets Supabase+GitHub. They coordinate via shared Supabase tables or Redis pub/sub.
What's the simplest subset to start with?
Tavily + Supabase + GitHub. Three MCPs, ~10 minutes to install, covers 70% of agent use cases (search, memory, code). Add Firecrawl when you hit 'I need structured content from a site', add Perplexity when you need citations, add Redis when the API bill starts hurting.