<aside> 🔌 The MCP servers I actually reach for. MCP (Model Context Protocol) is the bit that lets an AI assistant do things — search the web, send an email, query a database — instead of only talking about them. Below is the set I keep wired up, grouped by the job they do. A few are ones I built myself (marked ⭐). No keys or setup here — that's yours to add; this is the what and the when.

</aside>

🔎 Search & research

Brave Search — web, news and image search from inside the assistant, for current facts.

Perplexity — deeper, cited research runs for when one search isn't enough.

Context7 — up-to-date library and API docs, so the model stops guessing at old syntax.

qmd — BM25 + semantic search over my own markdown notes. The transferable idea: point the assistant at your knowledge base, not just the web.

🧠 Memory across sessions

claude-mem — searchable memory of past sessions and decisions, so the assistant isn't starting from nothing each morning.

💻 Run models locally (free and private)

⭐ local-llm (mine) — routes cheap, mechanical work to a model on my own machine. No API bill, nothing leaves the laptop. Good for classify / summarise / extract.

✉️ Newsletter, email & social

Resend — transactional and automated email.

• Kit — subscribers, sequences and broadcasts (this newsletter runs on it). Official hosted connector, not a repo.
• Buffer — drafts and schedules social posts. Official hosted connector.

📅 Comms & calendar

• Gmail · Google Calendar · Google Drive — triage mail, manage the calendar, reach files. Official Google connectors.
• Zoom — meeting recordings and transcripts. Official Zoom connector.

💳 Money & automation

Stripe — billing, invoices and payment links (the Agent Toolkit).

n8n-mcp — build, validate and fire workflow automations.