Command Center

flag Bottom line

You are on the canonical setup. Every tutorial in the YouTube/blog universe is showing the exact same thing you have: Private Integration Token (PIT) + locationId header โ†’ services.leadconnectorhq.com/mcp/. Same auth. Same endpoint. Same 21-tool surface. No one has a magic shortcut you're missing.

What makes it sound easier: they demo ONE location and end at "look, Claude listed my contacts!" You did this 10 times for 10 client sub-accounts, and the actual hard part โ€” the Circa playbook, governance, per-client context โ€” starts after connection, which is where their content stops.

The one genuine 2026 improvement worth adopting: per-folder .mcp.json scope instead of all-user-scope. Cleaner namespace, harder to send the wrong SMS to the wrong client. ~30 min to pilot on one client. See "Real opportunities" below.

check_circle What's actually wired today

Verified live via claude mcp list on 2026-05-23.

  • 10 GHL MCP connections, all โœ“ Connected, all HTTP, all pointed at GHL's native MCP server.
  • Auth method: Private Integration Token (PIT) + locationId header โ€” the only auth method GHL officially supports today (OAuth is roadmap).
  • Config scope: user-scoped in ~/.claude.json โ€” visible in every Claude Code session globally.
  • Tool surface: 36 tools ร— 10 clients = 360 tool routes available. Full catalog at /info/circa-ghl-mcp.
  • Specialist layer: Circa is the agent meant to drive these tools with a playbook (CLI-51 = build the playbook).

There is no version of this setup that's "more done" than what you have. The next step isn't more wiring โ€” it's the playbook layer.

view_list The 6 ways YouTubers / blogs connect Claude to GHL

When someone says "I'm using GHL MCP," they could mean any of these. Most of the "easier" claims come from conflating methods.

Method Topher Aria's read
1. GHL Official MCP + Claude Code โœ… Topher x10 Canonical. What Topher has.
2. Community MCP (mastanley13 / others) โš ๏ธ Alt More tools, but you host it on Vercel/Railway/Docker. Adds ops surface.
3. Claude Desktop Cowork Mode N/A Background scheduled tasks. Different UX, same MCP underneath.
4. Chrome extension (UI automation) N/A Visual scraping, not MCP. Fragile. Skip.
5. Windsor.ai connector on claude.ai N/A OAuth โ‰  writes. Wrapper service. Cancel-anytime SaaS.
6. Cursor / Windsurf / Gemini CLI + MCP N/A Same MCP, different IDE. No advantage over Claude Code.

visibility Why their setup LOOKS easier than yours

Five asymmetries between a 12-min YouTube demo and a 10-client agency reality.

1

They demo ONE location

Every tutorial walks through ONE PIT, ONE locationId, ONE config block paste. "10 minutes" is real for that. Topher did this 10 times. The architecture isn't harder; the keystrokes were 10x.

2

They stop at "Claude can see my CRM"

YouTube demos end at the "wow" moment โ€” Claude lists 3 contacts on screen. They never show Day 30, when you're asking "should this AI text my customer at 11pm without me approving?" That's the governance work no demo covers.

3

They conflate setup with leverage

Implicit pitch: connected = automated. Reality: connecting 10 sub-accounts is the easy part (done). The leverage lives in the playbook โ€” Circa โ€” that codifies which actions are direct, which need Topher's nod, what context each client gets. That's CLI-51, and YouTube content is silent on it.

4

They use single-account semantics throughout

"Your account / your Location ID" with no agency layer. Even tutorials titled "for Agency Owners" admit (in the body) that multi-tenant ops aren't actually explained. Single-tenant content is just easier to produce.

5

A few are showing different products entirely

Some YouTubers demo community MCPs (mastanley13's 269-tool self-hosted) or wrapper services (Windsor.ai OAuth, paid, read-only). When the label is "GHL MCP" but the product is a different SKU, it looks easier โ€” because it is, but it's also not the same thing.

fact_check Specific YouTuber / blog claims โ€” accurate?

Claim Accurate? Caveat
"Setup in 10 minutes, three steps" โœ“ True True for ONE location. Topher did this 10 times.
"15 minutes. Setup is fast once you have the right pieces" โœ“ True Same โ€” single-location demo.
"No middleware. No Zapier. No custom API wrappers" โœ“ True Correct. This is just the MCP value prop.
"Read contacts, update pipelines, send SMS, book appointments from plain English" โœ“ True Correct โ€” and this is exactly what Topher's 10 connections already do.
"Agency owners โ€” manage multiple sub-accounts" โœ— Misleading Implied in titles, NOT covered in the body. Multi-tenant ops barely addressed in any tutorial.
"OAuth-based, no token management" ~ Partial Real but READ-ONLY. Not the same product. Wrapper, not the official MCP server.

trending_up Real opportunities (the actually-useful takeaways)

Three concrete moves the research actually surfaces.

๐Ÿ†• Per-folder `.mcp.json` scope (genuine 2026 improvement)

Topher's 10 GHL connections are all USER-scoped in `~/.claude.json` โ€” visible globally, namespace-clutter in every session. Modern Claude Code pattern is one folder per client with its own `.mcp.json` (committable, isolated, auto-scoped when you `cd` in). This is what one tutorial calls out and it's a real ergonomic win.

Cost

~30 min to migrate, ~zero risk

Benefit

Eliminates 10ร— namespace pollution in every Claude session. Reduces "send the SMS to the wrong client" failure mode to near-zero. Sets the pattern for client #11.

Action

Pilot on 1 client (suggest Made By Aspire โ€” agency's own house). If clean, migrate the rest. Bundle into CLI-51.

โณ Watch the GHL roadmap (Hermes already doing this)

OAuth + npx package + 250 tools are GHL's stated roadmap. When OAuth ships, the PIT-management story (token rotation, secret hygiene, the doc-of-PITs on M4) gets meaningfully simpler. Hermes watch-loop armed 5/21 โ€” any GHL announcement bubbles to Aria's morning brief automatically.

Cost

zero (already running)

Benefit

Catches the next real shift the day it ships, not the day a YouTuber notices.

Action

Nothing to do. Status: Day 2 armed, no signals yet.

๐ŸŽฏ The actual leverage is CLI-51, not "more MCP"

The YouTubers are selling the wrapper-thin "I connected it" excitement. Topher already has the connections. What he doesn't yet have is the Circa playbook on top: governance, per-client context, the "Circa does X directly / asks Topher first for Y" rules. That's the gate, and it's already P1.

Cost

Not "more setup" โ€” it's the playbook work CLI-51 was created for.

Benefit

One Circa upgrade = leverage across all 10 clients. The thing YouTubers don't have.

Action

Ship CLI-51 internal proof-of-concept on the existing connections. The MCP layer is solved; the playbook layer is the work.

schema Where the YouTube content ends vs. where Topher's real work is

โ”Œโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”
โ”‚  YOUTUBE TUTORIAL SCOPE                                                  โ”‚
โ”‚                                                                          โ”‚
โ”‚   Make PIT in GHL  โ†’  Paste config  โ†’  Run `claude mcp list`  โ†’  Demo   โ”‚
โ”‚        (3 min)         (2 min)             (1 min)              (4 min) โ”‚
โ”‚                                                                          โ”‚
โ”‚                                                  END OF VIDEO  โœ‹        โ”‚
โ””โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”˜
                                                  โ”‚
                                                  โ–ผ
โ”Œโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”
โ”‚  WHERE TOPHER'S ACTUAL WORK STARTS                                       โ”‚
โ”‚                                                                          โ”‚
โ”‚  โ€ข Repeat ร—10 for each client sub-account                                โ”‚
โ”‚  โ€ข Per-client governance (Circa playbook โ€” CLI-51 "the gate")            โ”‚
โ”‚  โ€ข Which actions Circa does directly vs. confirms with Topher first      โ”‚
โ”‚  โ€ข Per-client context โ€” Patina โ‰  NRX โ‰  Modern Maid                       โ”‚
โ”‚  โ€ข Secret hygiene (10 PITs currently in a doc on M4)                     โ”‚
โ”‚  โ€ข Token rotation when GHL forces it (every 90d)                         โ”‚
โ”‚  โ€ข Per-folder .mcp.json migration (clean namespace, prevent cross-client โ”‚
โ”‚    accidents)                                                            โ”‚
โ”‚  โ€ข Catch the next real shift (OAuth, npx, 250 tools) โ†’ Hermes watch loop โ”‚
โ”‚                                                                          โ”‚
โ”‚                              โ† this is the leverage                      โ”‚
โ””โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”˜
        

link Sources surveyed

Related: /info/circa-ghl-mcp (operational reference for the 10 connections + tool catalog) ยท /info/aspire-digital (lane parent)