@elitedcs/ghl-mcp 3.94.1 → 3.95.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/CHANGELOG.md CHANGED
@@ -1,5 +1,39 @@
1
1
  # Changelog
2
2
 
3
+ ## 3.95.0: The brain checks a message against the lead's record before it goes out
4
+
5
+ **The brain can now check a message against the lead's record before it goes out.** You add one asking step in front of an SMS or email in a workflow. When a lead reaches it, the brain reads that lead's GoHighLevel record (contact, appointments, invoices and recent messages), its own notes on the lead, your business's path and your past rulings, and answers send or hold. If the brain cannot check it, a reminder still goes out and a message asking the lead for something is held for a person.
6
+
7
+ **Turn it on in three steps.** First, ask Claude to set your business's path for the judge: the steps a lead goes through, what each costs and what counts as done. Second, run the brain's wiring, which shows the ask key once. Keep it. Third, ask Claude to add the brain's step in front of the message, naming the workflow and the message step, and say whether it is a reminder. Claude shows you the workflow's full new step list first and changes nothing until you say yes. For the first brain step in an account, give Claude the ask key that wiring showed you once; after that the tool finds it on the existing brain step, so you do not paste it again. The tool can take the step out again while the workflow is unchanged.
8
+
9
+ **Every answer arrives within 6 seconds.** If anything is slower than that, the account's fail rule answers instead, and a reminder is still held when the brain or GoHighLevel already shows the lead opted out or on do-not-disturb. The brain records its judgment just after it answers. If that recording fails, the judgment shows as not judged.
10
+
11
+ **If the brain cannot be reached, reminders still go out and other messages are held.** A held message does not go out unless someone sends it by hand. A reminder is held only when the brain answers no; a message asking the lead for something goes out only when the brain answers yes.
12
+
13
+ **When the brain says not to send, its reason stays out of the workflow.** The answer says the brain is recording its reason and will try to add it to the contact as a note. The reason never appears in the workflow's run history.
14
+
15
+ **A wrong or replaced ask key holds every message.** A step still carrying an old key is held, reminders included, until it carries the current one.
16
+
17
+ **Know the limits before you rely on it.**
18
+ - For now it works only on accounts on this computer's sandbox list.
19
+ - A reminder the brain could not check goes out unless the brain already knows the lead opted out or is on do-not-disturb.
20
+ - It checks about 4 messages every 10 seconds per account. A workflow that sends to a whole list at once will have most messages decided by the fail rule: asks held for a person and reminders sent unchecked. A speed-up is planned.
21
+ - Anyone who can see the workflow's run history can see the ask key. Repeated fake requests could use up the judge's half of the day's GoHighLevel allowance. After that, reminders go out unchecked for the rest of the day. Drafts are not affected.
22
+
23
+ ## 3.94.2: A signed-out Claude Code is named as signed out, and a Build waiting on you says so
24
+
25
+ **Signed out means signed out.** If Claude Code is signed out on your computer, a step now says so and how to fix it: sign in to Claude Code again (type /login in Claude Code). Before, the start-up check read Claude Code's "Not logged in" answer as a sign that an update had changed something, and said so.
26
+
27
+ **The Queue says when a step's last run did not finish.** A step whose last run stopped before it finished now reads "the last run did not finish" in the Queue, with the reason from that run, instead of "Not run yet".
28
+
29
+ **A Build waiting for your confirm says it is waiting for you.** When Build stops at its list before writing anything, the Queue and the Board now say "Confirm what Build will do" and count it as waiting on you, and the step's status says Build stopped before writing anything. Before, the same moment read as a failed run, "Not run yet", and "Nothing is waiting on you".
30
+
31
+ **A re-run of the Brief starts clean.** While the Brief runs again, its card no longer shows the last run's message above it.
32
+
33
+ **Undoing an earlier build never leaves Build stuck.** When Build is held because an earlier build of a different plan is in the account, the step now has a Start the build button for after the undo. If that earlier build has been fully undone since the list was made, the step says the list is out of date instead of naming items that are gone. A step that Claude Code's sign-in stopped now says so: before it ran, that nothing in the account changed; partway through a step that writes, to review the account before running it again.
34
+
35
+ **Fixes on the client handoff.** Win-back's description now says it reaches leads who go cold from now on, which is what the build makes. On the account detail page, an automation built and left switched off is listed under its own heading, Built, switched off, instead of as "not in the account yet". The line about switched-off automations no longer mentions texting, which may not be part of the build. More handoff fixes are planned for the next update.
36
+
3
37
  ## 3.94.1: Fixes from a full walk-through of the cockpit
4
38
 
5
39
  **What they bought now comes before the Brief.** Building a client account is 13 steps: after the intake, step 6 asks what your client paid for, then the Brief and the Blueprint Review are written for exactly that. Before, you ticked what they bought at Approve, after the review was written, so the review was out of date on almost every build. If you change what they bought after the review, it says so in one line, and you choose: Approve anyway, or Update the review (this uses Claude). Approving anyway is recorded with the approval. Every step number the cockpit mentions follows the new order.
package/README.md CHANGED
@@ -904,7 +904,7 @@ Built by **[Elite DCs, LLC](https://elitedcs.com)** — a digital marketing and
904
904
 
905
905
  **Tech stack:** TypeScript, Node.js, esbuild, MCP SDK, Zod, GHL API v2, Firebase Auth
906
906
 
907
- **Version:** 3.94.1
907
+ **Version:** 3.95.0
908
908
 
909
909
  ---
910
910