Webinar: Giving an Agent SQL Access Without Getting Pwned

Mateo will show how to build an MCP server that gives any agent a real memory — backed by Postgres, append-only by design, and safe to expose to an LLM without it nuking your data.

We'll cover:
→ A schema for "agent memories" that supports edits and deletes without ever issuing UPDATE or DELETE
→ Parametrized write tools so the agent can save memories without touching SQL
→ Free text-to-SQL for reads — and the layered guardrails that make it safe (read-only roles, statement timeouts, row caps, multi-statement rejection)
→ Live attempt to DROP TABLE from inside the agent — and why it fails
→ Bundling skills with your MCP server so the agent actually knows how to use the tools

If you've ever built an agent that "kind of remembers things" with a JSON file and vibes, this is the upgrade path.

🔗 Try Arcade: https://arcade.dev?utm_source=youtube&utm_medium=video&utm_campaign=webinar_agent_memory_mcp&utm_content=description
🔗 Docs: https://docs.arcade.dev?utm_source=youtube&utm_medium=video&utm_campaign=webinar_agent_memory_mcp&utm_content=docs
🔗 Discord: https://discord.gg/arcade?utm_source=youtube&utm_medium=video&utm_campaign=webinar_agent_memory_mcp&utm_content=discord

#MCP #AIAgents #Postgres #Arcade

Видео Webinar: Giving an Agent SQL Access Without Getting Pwned канала Arcade