AI/ML

 Wondering if anyone else is running into the problem where output format of LLM Prompt response being very inconsistent and display the raw JOSN into the output field. We have been using the LLM action to generate case closure notes for a while now it has been working great until recently when we upgrade to 26r1. I understand that the AI Studio requires JSON output, but from my understanding, once the response reaches to the closure notes tab of the case screen, it should display in regular text in the format we specify in the instruction.Every time when I test the prompt, the preview tab shows exactly how I want to see it. But when I click on the LLM Action button within the case, I would sometimes get the raw json with no additional formatting. If I say in the instruction that I don’t want any json or code blocks, Acumatica would throw an error saying system couldn’t parse output as valid JSON. Has anyone else run into this? And if you haven’t, what does your instruction or output f

Hi everyone,I’d love to hear from the Acumatica developer community.Acumatica is evolving quickly with Modern UI, AI, APIs, automation, and cloud technologies. For developers who want to grow their careers, I’m curious:👉 What skills do you think an Acumatica Developer should focus on learning in 2026?Some areas I’m considering:🔹 Acumatica Modern UI 🔹 AI / AI Studio 🔹 C# & .NET 🔹 REST / SOAP APIs 🔹 Generic Inquiries 🔹 Acumatica customization 🔹 Performance optimization 🔹 Integrations with external applications 🔹 Azure & cloud technologies 🔹 Automation and workflowsFor experienced Acumatica developers:What skill has helped you the most in your career?For developers who are currently learning Acumatica:What are you focusing on right now?Would love to hear your experiences, recommendations, and learning resources.Let’s help each other grow as Acumatica developers! 🙌

Hi Acumatica Community!I’m releasing Acumatica Knowledge MCP, a free, read-only knowledge server for AI tools that support MCP.It gives AI assistants searchable Acumatica reference context across Help Wiki documentation, DACs and fields, OData entities, Contract-Based REST API entities, and Generic Inquiry XML examples.It does not require an Acumatica tenant connection. It is safe to use because it is read-only and does not write anything back to Acumatica.Setup is simple: add a single MCP URL to any agent or AI tool that supports MCP.Website with connection details and how it works:https://knowledge.acutools.netThis is the first public release, so feedback is very welcome. I’d especially appreciate hearing where the search results are useful, where they miss, and what Acumatica knowledge areas or tools you would like to see added next.I’m also happy to help anyone get it set up or answer questions.  

Hi everyone,I've created prompts in AI Automation that validate records (for example accounting entries).I'm currently writing the result to a UDF. However, I think the notes field of a record would be a better option. But that field isn't listed among the input fields.Have any of you dealt with this requirement before, and how did you implement it?

Check it out here: 

Good afternoon, We recently upgraded from 25R1 to 25R2, and upon trying to create the LLM connection through Anthropic for Claude, we are getting the following error when trying to test the connection:“The max_completion_tokens parameter has an invalid value. Specify the correct value for this parameter on the LLM Connection form.”  Has anyone else experienced this error and is there a solution to fix it there is no field on the LLM Connection form.  Our partner did ask Acumatica about it, but their response leaves a lot to desire and provides no solution. 

Hello, this is a message aimed at current Case Management customers,The CRM team is exploring a potential new AI feature for Acumatica CRM that automatically surfaces similar past cases, so support teams can find and locate solutions faster instead of solving the same problem from scratch every time. The goal is less manual digging, quicker resolutions, and more consistent answers across your team.To build this well, we want to test it against real-world data — so we're looking for a few community members with large, established case databases that make use of the Closure Notes functionality who'd be willing to work with us during early development. This would involve reviewing early versions and giving us direct feedback on relevance and usefulness of our findings.If you're interested (or just curious to learn more before committing), please reach me directly via DM — we'd love to chat. Thanks in advance to anyone willing to help shape this idea into reality. Best, The Acumatica CRM P

Hi all,We've open-sourced acumatica-claude-kit, a Claude Code plugin that packages the integration gotchas we've collected building against Acumatica's contract-based REST API and OData Generic Inquiries:https://github.com/hallboys/acumatica-claude-kitWhat's in it:Auth/session traps — why naive clients leak API license seats, License_LoginLimitExceeded, lockout causes, and the session model behind them. Silent failure modes — writes that return HTTP 200 but persist nothing, filters that return empty sets instead of errors, CannotOptimizeException, and how to detect each. GIQL / OData GI pitfalls — the v3-vs-v4 dialect split, datetime/UTC handling, parameterized-GI behavior over OData, empty-result diagnosis. Entity/action naming — the actual names for common contract-API operations (Release, ConfirmShipment, FinishCounting, etc.).Two design points that might interest other teams:Every entry is tagged UNIVERSAL vs TENANT, so platform truths are never mixed up with one instance's configu

For anyone new to the thread: MCP4Acumatica is an open-source (Apache-2.0) remote MCP server that connects Claude — and other MCP clients — to an Acumatica instance through the contract-based REST API and OData. It runs on Cloudflare Workers, each user logs in with their own Acumatica credentials (so their Acumatica role governs what the AI can access), sensitive fields are redacted, and every call is logged. It ships ~38 read-only entity lookups plus Generic Inquiry, schema-discovery, and utility tools. Repo: https://github.com/hallboys/MCP4AcumaticaHere's what's new since my last update (Previous thread).  Current Version 0.38.3Generic Inquiry governance (the headline)Exposing GIs to an AI assistant needs guardrails, so there's now an opt-in exposure gate (Note: This feature Needs a Customization and Couple of GIs):You choose which GIs the AI can see. Tag a GI with a new Exposed to MCP checkbox; only tagged GIs are discoverable/queryable. Until you configure it, the assistant can't b

Hi all,Happy Mothers Day!I needed a way to connect Claude to our Acumatica 2025 R2 instance right now, and the official Acumatica integration looks like it's still some time away. So I built one (entirely using Claude Code), ran it in production against our own ERP for a few weeks, and have now open-sourced it under Apache 2.0:Repo: https://github.com/hallboys/MCP4AcumaticaMCP4Acumatica is  a remote MCP (Model Context Protocol) server that runs on Cloudflare Workers. Each user logs in with their own Acumatica credentials — the server holds nothing centrally beyond an encrypted refresh token, and a user's normal Acumatica role governs what records they can read. Any MCP client (Claude.ai, Claude Desktop, Claude Code, ChatGPT) can talk to it.What's in the box (v0.31.1):38 read-only entity tools — Customer, Vendor, Sales Order, Invoice, Bill, Payment, Stock Item, Project, Case, Service Order, Shipment, Employee, and most of the common screens. 6 utility/discovery tools — generic-inquiry l

hello all,We have the AI Assistant feature from 26R1, however I would like to know if anyone has experimented it, and how?Thanks.