MCP Connectors by Databox - Give your AI Analyst context to explain performance and act
by•
Connect your AI Analyst to the tools your business runs on. It pulls context from your CRM or support desk, so every answer reflects what's happening in your business, and it can act on what it finds. Choose from 10+ connectors or add any MCP server.
Replies
Does it let users ask follow up questions after an AI Analyst response so they can dig deeper into the same performance issue?
Databox
@haolin Yes, Hao. It's a conversation, so you can keep asking questions and it keeps the context of the thread. For example, ask why revenue dropped, then follow up with "which accounts drove it?" or "what did the team say about those deals?" The AI Analyst goes back to your metrics and connected tools each time, so every answer builds on the last one.
Databox
It's worth pointing out that compared to a normal integration, this isnt just data coming in, its tools genie can actually call. It's basically the foundation for agents later, not a standalone feature on its own.
Databox
@marcel_mumel Exactly, Marcel. An integration brings data in. A connector gives Genie, our AI Analyst, tools it can use: it can look things up, and it can act, like updating a deal or creating a task. That's what makes it the base for agents. Every connector a team adds today is one more thing their agents will be able to do later.
But when it “acts on what it finds,” what are the limits? Can teams mandate that they get approval before it responds to the CRM or to the customer?
Additionally, Congratulations @zigapotocnik @pc4media and @rohanrecommends 🚀✌️
Databox
@aymi_malik Thanks, Muhammad! Great question. You stay in control of what it can do. Permissions are set tool by tool, with one main setting per connector and overrides for single actions. For example, you can let the AI Analyst update a deal stage on its own, but block it from sending anything to a customer.
Humalike
lfg!! congrats on the launch
Databox
@mcarmonas Thanks, Martí! Appreciate you stopping by.
Nice, congratz! I work in support and from a technical perspective it's awesome that every connector ships with read AND write tools. Write is just turned off until permissions and oversight catch up. This is a cleaner path to v2 than adding support for write access after the fact
Databox
@emil_korpar Thanks, Emil! Every connector ships with both read and write tools, and both are live. What keeps it safe is the permission setup. You choose what Genie, our AI Analyst, can do, tool by tool, so a team can allow updating a deal but block sending anything to a customer.
49agents IDE
Do you have mechanisms for provenance checking facts/numbers via some kind of multi-agent checks or "citing"? How do you make sure that there are no hallucinated results?
(i assumed that this might be the problem you have, but depending on structure of your soft ig you might even avoid this entirely with retrieval-based approach over generated (AI just needed to find and point to smth in the data instead of "read and then rewrite in the return response")
Congrats on the launch!!
Databox
@alp_adur Thanks, Latif, and great question. You're right that the retrieval side does most of the work here.
The numbers aren't generated. When Genie, our AI Analyst, answers a metric question, it queries your actual Databox data sources connected or uses correct MCP tool calls and works from what comes back. It doesn't recall or estimate numbers. Connectors work the same way. The context behind a number comes from live tool calls to your connected tools, so the AI Analyst is pointing at a real deal, ticket, or thread. It isn't writing its own version of what probably happened.
It also points you to the source it used, so you can open it and check. We don't claim hallucinations are impossible. No one building on LLMs honestly can. But grounding every answer in retrieved data and showing where it came from is how we keep it tight, and how you can catch it when something's off.
CopilotKit
Hey Peter, this is a great product. Congrats on the launch!
Databox
@nathan_tarbert Thanks, Nathan, appreciate your support!
Databox
Good job, team! Since launch, I managed to setup full end-2-end workflows, leveraging data retrieval via @Langchain MCP, semantic analysis of the conversations and digesting the results in the datasets. Meanwhile, I put agents to work on daily/weekly routines for insights and anomaly tracking. Unlimited data-analysis-related use cases.
Jakob, on the approval default: how does that work for a Routine running on a schedule, when nobody's in the chat to say OK? I run a few agents unattended on a Mac Mini and that's the case I'd worry about. A routine waiting on approval and a routine that found nothing to do look much the same from the outside, unless one of them says so.
Databox
Hey@dalemooney, a routine never waits for an OK. It only gets the MCP tools you've set to "always approve". Anything on "ask" or "not allowed" isn't passed to Genie at all, so it doesn't even know that tool exists and works with what it has. Routines are also instructed that nobody's watching: finish the job on your own, or end the run by saying why you couldn't get it 100% done. So "stuck" and "nothing to do" never look the same.
@grega_cej Hiding the tool rather than making it wait is neat, there's nothing to get stuck on. When a routine does end early, does its note say which held back tool it would have needed? That's the bit I'd want so I know what to switch on.
Macaly
#1 well deserved 🥇 giving the analyst context from crm + support is the usefull part. which connector gets used most?
Databox
@petrkovacik Thanks, Petr! It's only been a few days, so it's too early to give real numbers. So far, the CRM is the clear first pick for most teams, since that's where most of the "why" behind a revenue or pipeline change lives. Team chat and support tools come right after. Happy to share a proper breakdown once we have a few weeks of data.