Box AI API

I am trying to integrate IBM Box with IBM Bob IDE using MCP to query files and folders from Box in real time. Here is what I have done so far: Created a Platform App in Box Developer Console - Selected Server type with Client Credentials Grant - Configured mcp_settings.json in Bob.Issues I am facing:1. Package @box/box-ai-agent-toolkit-mcp returns 404 on npm — does not exist2. App status is Disabled — pending IBM enterprise admin approval3. Developer Token option is greyed outQuestions:1. What is the correct npm package name for Box MCP Server?2. Does Box MCP officially support IBM Bob IDE?3. How to get enterprise Box app approved faster?4. Is there an alternative way to connect Box to Bob without admin approval?Any guidance would be really helpful!

Has anyone else experienced the Box AI API Q&A endpoint returning 403 issues on all queries since last evening? It seems like the same API calls that were working yesterday are failing as forbidden, even though nothing changed with the App client’s permissions. Already verified the Client ID we are using has access to the file’s were using in the request but now only getting 403 back.

How often does Box AI update the default LLM model? How do we get advance notification that this will happen?

Installed Box for Salesforce and Box AI for Salesforce add on. Most of the calls work out of the box except a few such as Ask Box which would read a PDF stored in Box. Opened a Support case and the initial responses were incorrect.You can only access Box AI via Box Interface - No I just installed Box AI for Salesforce and it works. This is only available to Enterprise plans and up - Not according to this Oct 25 release from Box. https://support.box.com/hc/en-us/articles/45612941554835-Expanded-AI-API-Access-and-AI-Units-for-Business-Business-Plus-and-Enterprise-Plans So now I don’t trust the answers I’m getting. I just want a handful of credits to: 1. Confirm that the reason I’m getting the Unauthorization error for this specific API is because I need credits. 2. See if what it’s returning is what I can use for my Agent Action. The only purchase of credit option I see is for a full year?? I’d even buy a handful but I obviously do not want to commit to a year without trying. Support als

 These error was occor . Somebody Know?esponse_type=codeedirect_uri=https://vertexaisearch.cloud.google.com/oauth-redirectstate=eyJvcmlnaW4iOiJodHRwczovL2NvbnNvbGUuY2xvdWQuZ29vZ2xlLmNvbSIsInRpbWVzdGFtcCI6MTc3NTk3NTMwMTYyN30=client_id=ye0n4oqaz2qbh1j2ej1broyja7lixmv

I’ve spent the better part of the last month trying to give Box money for AI credits. Not negotiating, not asking for special pricing, just attempting to purchase what’s already advertised. I’ve been connected to multiple reps, each saying they’ll get back to me, and then… nothing.To be clear, Julianna has been great and has actively tried to escalate this on my behalf. The issue is that Joanna, who is supposedly the person responsible for actually enabling or providing the credits, has been completely unresponsive and unhelpful.This isn’t a feature request or a product complaint. It’s friction in the purchasing process itself. If a motivated customer actively trying to pay can’t get credits after weeks of follow-ups, that feels like something worth flagging.Curious if anyone else has run into this, or if there’s a clearer path I’m missing.

Most teams think they need custom model training to get real context. But in practice? Training is slow, expensive, brittle, and hard to govern. And for 95% of company use cases, you can get better results faster by pulling in your content at runtime using secure RAG, tailored instructions, and task-specific agents. In this episode of our AI Explainer Series with Box CTO, Ben Kus, we break down: When training actually makes sense (rare, high-risk, ultra-specialized workflows) Why most companies don’t need it — and how RAG + instructions deliver grounded, auditable answers instantly Real patterns from the field: caching, hybrid RAG, and policy-aware prompting to reduce latency, cost, and risk   Ben and I would love to hear from you…What is one task in your org that feels like it “needs” a custom model — and what problem are you trying to solve with it?  If you’re catching up, you can also check out other highly-requested explainers: Secure RAG for the Enterprise AI-powered Data Ex

 Hello, I'm trying to connect to the BOX MCP server. I referred to the guide here.  I couldnt get it working. How do I get the auth credentials?Is this endpoint correct?   I'm trying to connect to my our Agent Platform targeted at enterprise deployments  I wanted to try it with how we enabled shopify endpoints.  Best,Arun 

https://www.linkedin.com/posts/box_the-box-mcp-server-is-now-in-the-github-mcp-activity-7389044785325600768-GD3F/I just configured this and it is spectacular. All of our knowledge and best practices in one place. My team can simply use the ai_qa_hub tool to ask a question about our style and best practices. We have some specific SQL conventions we use and I had Github Copilot / Box MCP update the file. It came back with the specifics we needed. Exactly what I was looking for.I know other dev platforms allow you to put knowledge in their systems, but we have all of our other corporate knowledge in Box so this just keeps everything consistent and easy to reference.

I read the below aritcle. It’s awesome!!https://blog.box.com/box-community-mcp-server-updates-you-didnt-know-you-neededI have two questions:the authentication parameter is --box-auth, not --auth, right? I can’t connect n8n MCP Client that’s endpoint is set to http://0.0.0.0:8001/sse. I need set it to http:host.docker.internal:8001/sse. do you use a local docker and a local mcp server or not? If I set Authentication parameter “None”, the MCP Client can’t connect local MCP Server. the local MCP server displays the below error: WARNING:middleware:No token configured, rejecting all requests INFO:     127.0.0.1:53381 - "GET /sse HTTP/1.1" 401 Unauthorized   

I noticed that there is a change on Jun 24th, 2025 about Enhanced extract agent.Could you please tell me when I can use extract agent successfully in my code？(now when I post request to https://api.box.com/2.0/ai/extract I will get response with 400 bad request)Thank you for your support.

I’ve got a small Box AI application and we’re considering exposing the output of it to the public. Has anyone got some best practices for profanity and inappropriate content filtering with Box AI? It would be enough to just have Box AI say “I’m sorry. I can’t answer that question.” or something like that, when the user asks an inappropriate question. Do I just handle that through the prompt in the API call, or is there some other functionality I should be using that provides more structured content filtering? 

Where can I get the material &/or recording data by used the below workshop? 

Hi, we are using the AI API to extract structured metadata. The format of the response from the Box API keeps changing, sometimes multiple times a day. Sometimes it has the below structure, which shows the key name, the prompt, the type and value.   Answer: {fields: [ { key: financialYear, prompt: What financial year, if any, does this document relate to? Financial years run from 1 July to 30 June. Return the year only (for example the 2023/24 financial year would return 2024), type: string,  value: 2024       }     ]} And other times it simply returns the below:Answer: {financialYear: 2024} Every time it changes we are forced to update how the response is treated. Why would this keep changing? Can we force it to stay with the second example? Thanks. 

I need some help with the Box AI integration that pops up for each file I preview in my Box folders.(Sorry in advance for any inaccurate language. I’m an end user biologist, and information technology is largely beyond me.)I’m not going to use this integration. Full stop. Period. Putting aside the ethical and environmental concerns that I have about generative AI, I do not trust the application to be accurate with my information. I work in research, so information in the documents that I generate are still likely speculative and/or unproven and the information that is available might not be peer reviewed or validated.That being said, I find it incredibly annoying that the first thing that happens every time I preview a document is I’m prompted to start a conversation with Box AI. Is there a way to disable this integration on the users side? It’s galling that I’m being prompted again and again to use a tool that I think will be actively detrimental to my work and ability to utilize Box.

What type of permissions one has to have in order to get a User Activity Report on a file that they own? I can’t obtain a User Activity Report at all. I’m part of an Enterprise + Plan.

 We’ve just returned from an exciting BoxWorks conference where we demonstrated groundbreaking AI integrations between Box and HubSpot that are set to transform how businesses handle document processing and customer relationships.Automated PDF Data ExtractionOne of our most compelling demonstrations showcased Box AI’s ability to automatically extract and process metadata from PDFs stored in Box. When triggered by a HubSpot action, the system can intelligently parse this data and update corresponding HubSpot fields automatically. This innovation eliminates manual data entry, reducing both time investment and the risk of human error.Think about the implications:Automated contract processing Streamlined workflow triggers Seamless data synchronization between platforms Reduced administrative overheadCross-Platform IntelligenceWe also unveiled a powerful new capability that leverages Large Language Models (LLMs) across both Box and HubSpot platforms. This integration enables users to make n

We have a new blog post to checkout on our Medium site. This demo goes over how to connect Box & Pinecone. https://medium.com/box-developer-blog/demo-box-pinecone-f03783c412bb Please check it out! 

Checkout the final video in the Box AI Platform API YouTube series between Box developer advocate Alex Novotny and VP of developer relations Laurie Voss.  

Lots happening in this space, checkout our recent roundtable recording on Box AI APIs with @shurrey Box which also touches on AI developer roadmap and a Reference Architecture which is being developed. 0 mins - Introduction1 mins 30 secs - Overview of Box AI5 mins - Box AI use cases7 mins 14 secs - Intro to Box AI Platform API14 mins 07 secs - Introduction to Box for AI18 mins - customer Q&A23 mins 15 secs - Overview of Box for AI - LangChain and Llama-Index26 mins - Box AI developer roadmap33 mins 30 secs - Customer Roundtable and Q&A48 mins 26 secs - Wrap up 👉🏽 You can also follow our Developer Blog here, to keep track of future updates. 

Hey Everyone!  Please checkout our YouTube playlist on all things Box AI Platform API, including demos, overviews, and interviews! Alex 

I heard there may be the ability to connect my own LLM to Box via API.  Is that on the roadmap or a known feature coming?  

What types of user interactions with Box AI are recorded in the User Activity Report, and how can Box Admins utilize this information?

Do you have questions about the Box AI API? This is the place to ask!