← Back to codykeegan.ai

MCP for operators

What it is, why it matters, and three workflows you can build today · By Cody Keegan
The short version: MCP is the standard way AI assistants connect to the apps you actually use — your CRM, your accounting, your calendar, your inbox. Before MCP, every connection was a custom integration. After MCP, an AI can talk to any tool that supports it without a developer.

What MCP actually is

MCP stands for Model Context Protocol. Anthropic released it in late 2024 as an open standard. The idea: instead of every AI tool needing custom code to talk to every other tool, MCP gives them a shared language. Like USB-C for AI.

Before MCP, if you wanted an AI assistant to read your inbox, update your CRM, and post to your calendar, you needed three custom integrations — each one written by a developer or assembled in Zapier with a brittle chain of triggers. After MCP, the AI assistant calls each app directly through a shared interface. Same protocol. No custom code.

You don't need to know any of the technical details. You only need to know one thing: MCP is the reason your AI assistant can finally do things in your business, not just talk about them.

Why operators should care

Here is the operator translation. Run through the Operator Filter:

That sounds small. It is not small. It is the difference between "AI as a writer" and "AI as an employee."

Which apps support MCP right now

The list is growing fast. As of May 2026, the operator-relevant ones include:

AppWhat you can do
Google WorkspaceRead inbox, draft and schedule emails, create calendar events, search Drive
HubSpotRead and update contacts, deals, pipelines; log calls and notes
QuickBooksPull P&L, reconcile transactions, flag discrepancies
PayPal · Stripe · SquarePull recent sales, refund history, invoice status
SlackRead channels, post messages, summarize threads
Notion · Google DriveRead and write documents, search the workspace
DocuSignSend contracts, check signing status
GoHighLevelPull leads, update pipelines, trigger automations mid-call

Three workflows I'd build today using MCP

1. The end-of-day inbox triage

Every evening, your AI assistant reads the day's inbox, drafts replies to anything that needs one, creates HubSpot tasks for leads that came in cold, and posts a one-line summary to your Slack saying "you have 4 replies waiting for approval — here's the gist of each."

Tools: Gmail MCP + HubSpot MCP + Slack MCP. Build time: under an hour. Cost: whatever your AI assistant subscription is. Operator value: 30-60 min/day reclaimed.

2. The monthly close packet

On the 1st of every month, your assistant reads QuickBooks, pulls the prior month's P&L, reconciles it against PayPal and Stripe settlements, flags anything that doesn't match, writes a plain-English narrative explaining what changed, and emails the packet to your accountant.

Tools: QuickBooks MCP + PayPal MCP + Gmail MCP. Build time: a couple hours. Operator value: 2-4 hours/month, plus a much cleaner handoff to your CPA.

3. The cold-call follow-up loop

Voice agent (Vapi) takes the inbound call, qualifies the lead, books the appointment. After the call ends, MCP triggers: HubSpot logs the new deal, GoHighLevel kicks off the follow-up sequence, Google Calendar holds the slot, and Gmail queues the confirmation email. Zero human touch from ring to confirmation.

Tools: Vapi + GoHighLevel MCP + Gmail MCP + Calendar MCP. Build time: a long Saturday. Operator value: every missed call after hours captured cleanly.

Operator Filter take: I've watched a hundred "AI is going to change everything" announcements over the years. Most of them didn't. MCP is the one I'd actually bet on, because it's solving a real plumbing problem: the integrations between apps were always the bottleneck. MCP isn't sexy. It's just where the money lane opens.

How to start using MCP

  1. Use an AI assistant that supports MCP. Claude (Anthropic), ChatGPT (with custom GPT integrations), Cursor, and most major coding assistants now do.
  2. Pick one app to connect first. Gmail or QuickBooks are the highest-leverage starts. Both have official MCP servers.
  3. Connect it through your assistant's settings. Most assistants have a "connect an app" or "add a connector" option. Authenticate. Done.
  4. Run one workflow. Don't try to automate everything on day one. Pick the most painful repetitive task in your business and build that single workflow. Iterate from there.

What MCP is not

MCP is not magic. It is a connection standard. The AI still needs to know what to do once connected. Garbage prompts produce garbage workflows.

MCP is not free agency. The AI doesn't decide on its own to send 500 emails. It does what you tell it. Keep human-in-the-loop on anything that touches money, customers, or public posts.

MCP is not finished. The standard is evolving fast. Tools that support it today might support it differently in six months. Build patterns, not custom code that breaks when an MCP server updates.

The full blueprint with my live MCP stack

I publish a new operator-flavored AI blueprint every Saturday. The next one is a complete MCP-driven workflow for a service business. Subscribers get it first.

Related reading

Last updated May 27, 2026 · © Darkhorse Traders LLC