Skip to main content
J
john.rogers
Employee
August 7, 2026

Discussion and Feedback on Dev MCP

  • August 7, 2026
  • 6 replies
  • 119 views

Have questions or feedback on the Appian Dev MCP? This is a great place to ask them. 

 

Appian’s Dev MCP plugin can be found on the Appian App Market here (https://appmarket.appian.com/listings/appian-dev-mcp), as well as a read-only version intended for production use or other environments where it’s important to avoid any design object changes (https://appmarket.appian.com/listings/appian-dev-mcp-read-only). 

 

Documentation and installation guidance can be found here: https://docs.appian.com/suite/help/latest/devmcp.html

And finally, our public repo of skills to install alongside the MCP can be found here. We welcome contributions if you’ve developed a skill or skill update you think would benefit other users. https://github.com/appian/dev-mcp-skills

    6 replies

    J
    john.rogersAuthor
    Employee
    August 7, 2026

    The next release of the plugin (26.6.85) is coming early next week. Here’s a sneak peek at the release notes:

     

    New tools: view object version history and past versions

    New execution behavior: Locked objects now reject edit attempts with a clear error

    New diagnostics tool for troubleshooting and getting version info for bug reports

     

    Tool Fixes

    • Document and folder descriptions are viewable
    • Changing document metadata no longer versions the document
    • Changing folder metadata no longer results in a new version
    • Fixed record type deletion feedback for agents
    • Fixed swimlane assignment and serialization in process models
    • Adding a new site page no longer causes existing site pages to be recreated and lose configurations
    • Date, datetime, and time types now function reliably for constant values and for testing inputs
    • Application creation works for non-admin Designer users
    • Record types and web APIs can be edited immediately after creation for non-admin Designer users
    • testProcessModel correctly handles record-typed parameters

     

    Auth and Ops

    • MCP server now auto-trusts corporate certs (users no longer get SSL errors on corporate networks)
    • Users no longer need to set LCP_API_PATH in their config
    • Multi-user auth improvements (users prompted less frequently)
      stefanhelzle0001
      stefanhelzle0001
      Brainy
      August 8, 2026

      Thank you so much for the plugin and allowing us feedback. We found that translation sets seem to not be supported. Do you plan to add support?

        J
        john.rogersAuthor
        Employee
        August 10, 2026

        Yep! We expect to get to all of them over time. Our near-term focus has been on deeper coverage on the most commonly-used objects, as well as some experiments to see what kinds of testing tools most expand the amount and quality of work the AI can do before pausing. Both of those have been more of a current focus than expanding to new objects.

        With that said, we were just discussing Friday which objects needed support next. Portals, translation sets, and task reports (and/or skills support for working with tasks) all came up near the top of the list.

          stefanhelzle0001
          stefanhelzle0001
          Brainy
          August 10, 2026

          Cool. Thanks. When do you plan a 100% coverage of all objects?

          M
          mikep0007
          Known Participant
          August 20, 2026

          I have been using this extensively the last week as I am trying to create a sales demo.  In this process i have found a few issues.

          1 - AI Skills.  DevAgent when it modifies them, actually messes them up and I need to re-create them.  It changed the input parameters and needless to say the AI skill does not work after that.  The visualization in the process model is also incorrect.  Unsure how/why it does that, but it is an issue

          2 - Given #1, I asked it to fix some json output from the AI Skill but do not modify the AI skill.  IT modified the AI skill anyway.  It also thought there was an error in the AI Skill so ti tried to fix it and back to #1

          3 - Many times it tried to fix a process model and gets stuck trying to fix it.  It spent 7 hours tryiung to fix an issue.  It should stop or ask for some guidance.  But within Claude, using the DevMCP, it spent that time re-fixing the same error over and over./  Fixed for a “known” issue.  that caused the same problem so it went back to the previous code and just kept spinning.  I stopped it and had to redo the entire task as it did not work at all.

          4 - more interesting type feedback, when working with record types in the code, it does not always work.  IT uses UUID and then in the code it does not reconcile correctly so we need to re-code the exact same thing.  And when you do see the real record name listed it is not understood, even though it looks correct.  These issues are not consistent so I cannot explain any pattern.

           

          The composer fix was a biggy and that also apparently fixed the Dev Agent.  I have found the Dev Agent o behave better than the devMCP in its ability to get it right he first time.  I would assume there is additional context in that agent that the MCP/Claude does not have which makes sense.

          Terms & ConditionsAccessibility statement