Skip to content

Latest commit

 

History

History
71 lines (41 loc) · 2.04 KB

File metadata and controls

71 lines (41 loc) · 2.04 KB

edgepilot - speaking script

Notes: female AI voice, conversational pace, pauses marked [PAUSE] Target: 90 seconds narration, 3 minutes total with demo running


This is EdgePilot.

Its two AI agents using the StackQL MCP server interacting with Cloudflare and Confluent from a single SQL interface.

[PAUSE - show repo in editor or terminal]

Before we run the agents, we provisioned the Kafka topic using stackql-deploy. One command. No state file. If the topic already exists, it does nothing.

[PAUSE - show terminal with stackql-deploy output, or just skip to demo.py]

Now let's run the demo.

[RUN python demo.py - PAUSE 3-4 seconds while MCP server starts and tools load]

The MCP server is live. Those are StackQL tools - Cloudflare, Confluent, loaded from a single local server.

[PAUSE - recon agent's first tool call appears]

The recon agent just ran a SELECT against live Cloudflare analytics. Not a cached state file. The actual API, right now.

[PAUSE - second tool call, rate limit query]

Now it's reading the current rate limit config.

[PAUSE - recon result prints]

Traffic is elevated. Twenty-three threats in the last thirty minutes. The recon agent hands that off.

[PAUSE - action agent starts, first SELECT visible]

The action agent reads the same rate limit config to confirm what it's changing.

[PAUSE - UPDATE visible]

Tightens the threshold. One SQL UPDATE against Cloudflare.

[PAUSE - INSERT into Confluent visible]

And now that INSERT is going to a completely different cloud - Confluent Kafka. Same MCP server. Same SQL interface.

[PAUSE - "done" prints]

That's it. Two agents. Four SQL statements. Cloudflare and Confluent from one query layer.

Infrastructure as data.

[END]


Claude Desktop version (optional follow-on, 30 seconds):

The same MCP server is wired into Claude Desktop. So you can run this interactively - ask it to check your Cloudflare zone, tighten a rule, log the decision - and it uses the exact same StackQL tools, the exact same SQL, no code required.

[SHOW Claude Desktop with a natural language prompt executing]