Most people using AI do this: Install a Claude connector, hit a wall & assume th
By John Aspinall ✱ · May 16, 2026 · Curated by George's Blog
Most people using AI do this:
Install a Claude connector, hit a wall & assume that's the ceiling.
→ It isn't.
Ran Week 2 of the Operator Intelligence cohort last night and the most valuable 20 minutes wasn't the skill we built - it was what happened when the skill broke.
The native Slack connector won't post to channels with external members.
Hard limit.
So we ran the workflow:
→ Searched GitHub for a third-party MCP
→ Validated it in Perplexity against the exact failure mode
→ Brought it back to Claude in plan mode
→ Asked Claude to confirm it solves the problem before touching anything
→ Installed, OAuth'd, stored the token in .env
→ Re-ran the skill. It worked.
That loop is the unlock.
The native connector library is the front door.
It's clean, fast & gets you 80% of the way.
The other 20% - the part that actually separates operators from people watching demos on YouTube - is knowing what to do when the stock tool can't do what you need.
⊗ You don't stop.
☑ You find the MCP.
☑ You validate it.
☑ You install it.
➅ minutes later you're doing things the platform technically doesn't support out of the box.
This is the kind of thing we work through live in OI.
☒ Not theory.
☒ Not "here's what's possible."
Actual builds with actual breakage & actual workarounds.
June 2026 Cohort details coming soon.
⤷ DM me if interested.