I've started hating gummie now :(

Is it just me, or has Gummie become increasingly dumb-er? I remember making pretty decent use of Gummie. However, since the past week or so, it seems to have lost the ability to help with anything useful- even basic stuff. It forgot that MCP nodes are now possible with Gumloop and said it random things like the feature hasn’t been launched yet. Feel like it has regressed accidentally to the first version when it was completely daft :frowning:

Hey @kritikakapur! If you’re reporting an issue with a flow or an error in a run, please include the run link and make sure it’s shareable so we can take a look.

  1. Find your run link on the history page. Format: https://www.gumloop.com/pipeline?run_id={{your_run_id}}&workbook_id={{workbook_id}}

  2. Make it shareable by clicking “Share” → ‘Anyone with the link can view’ in the top-left corner of the flow screen.
    GIF guide

  3. Provide details about the issue—more context helps us troubleshoot faster.

You can find your run history here: https://www.gumloop.com/history

Hey @kritikakapur – You should have received an email from Max, we’d love to hop on a call, gather feedback and make things work for you.

Hi @Wasay-Gumloop, I am a Gumloop newbie but pretty experienced with automation platforms and agentic workflows. I was pretty excited to get started in Gumloop but have to unfortunately share @kritikakapur frustration. I am specifically commenting on the Gmail MCP node - I’ve tried to make it work in various ways but end up just missing the mark. It doesn’t seem to understand how to use its own tools correctly and overriding this with better code is near to impossible they way it is currently set up. I’d be open to walk someone through the flow and steps I’ve taken to make it work.

Hey @Andreas_Oszkiel – Sorry to hear that! Can you share exactly what you were trying to achieve with the Gmail MCP? I can try to reproduce and fix the issue ASAP!

Also I’m happy to hop on a quick call to troubleshoot this: https://cal.com/wasay-ahmed/15min

Thanks for following up @Wasay-Gumloop. I have booked a call for next Tuesday. I’ve shared the flow link in the calendar booking.

Here is what I wanted to accomplish.

:e_mail: Step 1: Email Detection

  • Gmail Reader monitors your inbox for new emails
  • Scans for emails with attachments
  • Captures email metadata (Message ID, sender, subject, etc.)

:robot: Step 2: Invoice Classification

  • Ask AI (wrapped in IF/ELSE) analyzes email content and attachments
  • Determines if the email contains an invoice or receipt
  • Uses intelligent prompts to identify invoice-related keywords
  • Decision Point: Only proceeds if classified as invoice

:memo: Step 3: Filename Generation

  • Suggested Filename node extracts or generates appropriate filename
  • Creates structured naming convention for the invoice file
  • Ensures consistent file organization

:cloud: Step 4: Google Drive Upload

  • Upload to Google Drive saves the invoice attachment
  • Files organized in designated folder structure
  • Maintains original file format (PDF, etc.)
  • Provides shareable link for future access

:incoming_envelope: Step 5: Email Management ← this simple step caused the frustration, because it seems that it could only be accomplished with the Gmail MCP node

  • Gmail Mark Read & Archive processes the original email
  • Removes from inbox (archives to All Mail)
  • Marks as read

:bar_chart: End Result:

  • :white_check_mark: Invoice safely stored in Google Drive with organized naming
  • :white_check_mark: Email archived and marked as read
  • :white_check_mark: Inbox stays clean - no manual intervention needed
  • :white_check_mark: Fully automated - runs continuously in background
  • :white_check_mark: Only processes actual invoices - ignores other emails

This is my issue with the Gmail MCP node:
Setting this node up was the most time consuming. It is unclear how the MCP tools have to be queried properly and what their capabilities are exactly.
I must have tried to make this node work 20 times or so. In the end I settled for a mix of Claude and Cursor to edit the code of this node to get it to work. But this experience was quite frustrating because it was unclear which tool(s) to query to accomplish the tasks and also how to query the tools. The process of describing in natural language what I want from this node, only for it to come up with unusable code, then to go through multiple steps to edit the code, test the code, and run the flow to diagnose what goes wrong exactly felt very inefficient because the generative AI approach gets in the way every single time. Don’t get me wrong, I think having an MCP node is a fantastic idea. Also the option to use generative AI is great. But there should be an alternative also.
I couldn’t find any documentation (and I don’t mean the general documentation of the principles of how MCP nodes work). I was looking for an explanation of the tools you have preconfigured for this Gmail MCP node, the capabilities, how to best prompt the LLM to come up with the right code, what input and format the tools expect, etc. otherwise it’s nearly impossible to diagnose where it goes wrong.
This is probably applicable to all MCP nodes with generative AI to create the code.
I hope this helps.
Cheers
Andreas

also noticed that the name label of the custom MCP node isn’t updating properly in the list of available nodes.

Really appreciate the feedback and thank you for taking out time to provide that, we’ll work on this!

also noticed that the name label of the custom MCP node isn’t updating properly in the list of available nodes.

Fixed the edge-case here, should be live by Monday/Tuesday.

This topic was automatically closed 5 days after the last reply. New replies are no longer allowed.