Introducing Visdum MCP : Ask Your Comp Data in Claude

A few weeks ago, a compensation admin at one of our customers spent two full days on a single payout dispute. One rep, one deal, one number that looked wrong.

The plan was correct. The data was correct. The calculation was correct. But nobody could explain it fast enough to make the rep believe it. So it escalated. Finance got looped in, and a deal that closed in March turned into a debate that dragged into April.

Here is the part that stayed with me. She did not need a better calculation engine. Visdum already did the math perfectly. She needed something that could look at the plan, the deal, the approvals, and the history, and then say, in plain language, why this rep earned this amount.

That gap is the reason we built what we are announcing today.

Key takeaways

• Visdum has launched Visdum MCP, a governed platform that lets AI work directly on your sales compensation data. It sits between AI and your incentive system, so AI can read and reason over your plans, credits, and payouts, but only inside each user's permissions and always with an audit trail.
  
  
• It's built to answer "why," not just calculate. Older comp software tells you what a payout is; Visdum MCP explains why a rep earned it, tracing the number back to the exact plan rule and deal, in plain language you can hand to Finance or a disputing rep.
  
  
• It's the governed foundation under Visdum's AI layer, and it's live now. The AI Copilot is already shipped, the Report Builder, Nudge generator and a new connector that brings your comp data into Claude are the latest to join them, and because every capability shares the same base, trust and security get solved once rather than rebuilt for each feature.

Automation answered "what." Intelligence has to answer "why."

For a decade, sales compensation software has been an automation story. Ingest the data, apply the rules, produce the payout, retire the spreadsheets. That was the right fight, and it moved our customers a long way: go-lives measured in weeks rather than the multi-month norm, and hundreds of Professional Services hours taken off their plate.

But automation only ever answers one question well: what happened?

The questions that actually consume a compensation team's week sit underneath that one. 
Why did this rep's commission drop this quarter? Can I trust this payout run before I release it? What happens to margin if we move this accelerator? Can you just build me the report instead of me filing a ticket?

Answering those takes more than a calculation engine. It takes something that can reason over your plans, explain a number, and stand behind it. That is a different category of product.

Why "add AI" was never going to be enough

Most AI in compensation software is a chat box bolted onto a dashboard. It answers questions nobody was asking, on data it half understands. We held ours to a harder test: would a controller trust it on a live payout run?

Point a large language model at an enterprise incentive system with no guardrails and it will hand you confident answers you cannot verify, on data it should never have touched. In a system where every number lands in a paycheck and an auditor's file, "usually right" is worse than useless.

So the intelligence was never the hard part. Modern models already reason well. The hard part is giving that intelligence secure, governed, explainable access to the business logic underneath, without ever letting it guess, overreach, or act outside the rules.

That foundation is what we built. We call it Visdum MCP.

What Visdum MCP actually is

Visdum MCP is the trusted layer that sits between AI and your incentive system. MCP stands for Model Context Protocol, an open standard for connecting AI to external systems. That is the plumbing. What matters is what it does.

‍
It gives every AI capability in Visdum one shared, governed understanding of your world: the plans, the commission logic, the rate tables, the transactions, the credits and adjustments, the org hierarchy, the historical payouts, the reports and statements, and the permissions that decide who sees what.

Instead of each AI feature reinventing that logic on its own and drifting out of sync, they all stand on the same foundation. That is what makes them consistent, reliable, and safe to run on a payout. And it rests on five things a finance or RevOps team cannot compromise on.

Secure access. AI reaches your compensation logic through a governed gateway, never a back door, with authentication, authorization, and auditing built in. No shared API key, no service account.

Enterprise permissions. Each person signs in as themselves and sees exactly what they are allowed to see, not one row more. Turning AI on never widens who can see what.

Explainable reasoning. Every answer can be traced back to the plan clause, the deal, and the rule that produced it. No black boxes on numbers that end up in a paycheck.

Governed actions. The AI operates inside defined boundaries. Almost everything it can touch is read-only, and anything that writes waits for a human to approve.

Complete auditability. Every question asked and every answer given is logged. When Finance or your auditor asks why the system said that, there is a record.

Strip those away and you have a party trick. Build them in from the base and you have a platform that AI can stand on safely.

TL;DR: Visdum MCP is not a feature you switch on. It is the governed foundation every AI capability in Visdum reads from, so trust, permissions, explainability, and audit are solved once and inherited everywhere.

Visdum MCP for Claude

This is the same governed foundation turned outward. 

Connect Visdum to Claude and your team can ask about their own live compensation data in plain English, questions like "which deals were credited to EMEA in Q2?" or "why is this rep's payout lower than last quarter?"

The connector opens up 23 business datasets spanning the full compensation lifecycle, plus your own configured data feeds and rate tables. 

What can customers do with it?

The connector shows each person only what their own Visdum login already allows, so the same question surfaces different detail depending on who is asking. A rough map of who uses it for what:

How easy is it to adopt?

There is no rollout to schedule. Most of the adoption questions a Finance or IT team raises have short answers:

Governed access, checked continuously 

The concern is legitimate and well documented. Cisco has warned that the rapid spread of MCP has opened a large and often unmonitored attack surface, because AI agents can now reach databases and act on a person's behalf, and many organizations are not applying the security rigor they would give any other system with that level of access. 

That is exactly the failure mode we designed against. Each person signs in as themselves, and access is checked continuously, not just once at setup.

The short version: it reads, it cannot change anything without explicit approval, and each person sees only what their own Visdum login already allows.

TL;DR: Turning on AI does not widen access. Same identities, same permissions, same audit trail, reached through a governed gateway that fails closed.

Why solving it once, at the base, changes what comes next

Every capability runs through the same governed foundation, so trust, permissions, explainability, and audit get solved once and inherited by everything above. Less logic gets rebuilt per feature, and the platform gets more reliable as it grows, not more fragile.

That is what turns today's release into a starting line rather than a finish. The platform that explains a commission today is what will let AI safely build the plan tomorrow. On the same foundation, here is what comes next:

• Plan Builder: turns your plan documents and historical models into draft Visdum configurations, ready for review.
• Forecaster: compares plans and simulates payout cost before rollout.
• Agentic workflows: carry a run from calculation to approval, with continuous payout monitoring and automated plan optimization.
• Partner copilots: built by your own teams and partners on the platform.

None of that is reachable by adding a chat box to a dashboard. All of it is reachable once the trusted layer sits underneath.

Where Visdum MCP fits in Visdum's AI suite

Visdum MCP is the foundation. The assistants that sit on it are what your team touches day to day, and because they all read the same live plans, credits, and payouts, each one removes a different slice of manual work without you re-explaining your setup. A quick tour of the rest:

• AI Copilot - Ask anything across plans, credits, calculations, and payouts, and it explains how the result was reached - right down to tracing a broken payout to the one link that failed. One missing credit, four downstream effects: no component picks it up, the calculation returns zero, nothing gets approved, nothing pays out.  The Copilot walks that chain backward and names the break. And when it can't find an answer, it says so instead of guessing. The two-day dispute I opened with becomes a two-minute answer.
  

• AI Report Builder - Say what you want the way you'd say it to a teammate - a monthly payroll summary, a territory payout, a finance accrual - and watch it render live from your actual figures, not a mockup. Refine it in the same chat, then save it to your reports list to re-run any period. The line is clear: it reads and presents your numbers, it does not change a plan, a rate, or a payout.
  
  
• AI Nudge Generator -  Pick a rep and AI drafts a personal performance email from their real position - attainment, component breakdown, gap to target, one coaching tip. Every nudge is a draft first. You read it, edit it, and decide whether it sends. It drafts, you approve; nothing reaches a rep without your sign-off.

The point is not a handful of separate tools. It is one system that reads your compensation data and does the work around it, whether that work is answering a question, pulling a report, coaching a rep, or reaching your data from inside Claude.

Customers rate that system-level approach: Visdum ranks #1 of 20 for Customer Relationships in Mid-Market sales compensation, scoring 9.81 (G2, Summer 2026).

The bigger picture: a System of Intelligence for Revenue Incentives

Every enterprise already has its systems of record. CRM for customers, ERP for finance, HRIS for people. They store what is true.

None of them reason over your incentive system: explain it, optimize it, and act on it under your governance. That is the layer we are building Visdum into. Humans define the objectives, AI agents do the operational work, and Visdum MCP is the trusted platform that connects the two. That advantage is durable, and it reaches well past commission calculation.

The foundation is new. The track record under it is not. Mid-market teams already rank Visdum at the top of G2 for ease of use, ease of setup, estimated ROI, and the relationships behind the software.

It shows up in the exact problem this post opened with. Cyble runs its commissions on Visdum, and CFO Ernest Fung describes what happened to the disputes:

"There are very little disputes with the sales team on what the right level of compensation is. We provide full clarity and transparency to them. I think overall the sales team is happier." - Ernest Fung, CFO, Cyble

Conclusion

Confidence is believing your compensation system got the math right. Accuracy is believing your team can understand, trust, and act on that math the moment they need to, and soon, letting AI do the acting for them, safely, inside your rules.

Most of the market is still adding AI to compensation software. We built the platform that lets AI operate on it, securely, explainably, and under your governance. Not another feature. The foundation every current and future AI capability in Visdum stands on.

The automation era answered what happened. This is where Visdum starts answering why, what next, and can you do it for me.

See it answer against your own numbers

The fastest way to feel the difference is to watch it work on a plan you actually run. Bring your plans, your data, and your questions, and we will walk you through the Copilot, the Report Builder, and the Claude connector on a real Visdum account, using your business scenarios rather than a canned demo.

About The Author

Sameer Sinha is the Co-founder and Head of Product at Visdum, a leading Sales commission platform for sales compensation automation used by mid-market and enterprise companies. With over two decades of experience across enterprises like Oracle and Citibank, as well as scaling startups, he has deep expertise in enterprise sales, finance, and RevOps, and brings a practitioner’s perspective to solving complex commission challenges. He works closely with Finance, Sales, and Operations leaders to design scalable compensation programs, streamline data integrations across CRM, ERP, and HRIS systems, and drive transparency in incentive payouts. His insights span commission plan design, quota and territory planning, payout governance, and system architecture for high-scale environments. Sameer regularly writes about sales compensation strategy, SaaS integrations, and AI-driven revenue operations systems, helping organizations move from spreadsheet-driven processes to automated, audit-ready platforms.

FAQs

What is Visdum MCP?

Visdum MCP is the governed platform layer that lets AI read, explain, and reason over your sales compensation data. It connects AI to your plans, rules, rate tables, transactions, payouts, and permissions through a secure gateway, so every answer stays inside a user's access and leaves an audit trail.

Does connecting Claude give it access to our whole database?

No. Only an approved set of business datasets is reachable, and only for reading. The AI never touches the underlying databases directly, and queries are restricted to an approved list with limits on how much any single answer returns.

Can one user see another user's or another customer's data?

No. Each person signs in with their own Visdum credentials, and their access is tied to that identity. Your organization is resolved on Visdum's server from the signed-in user and cannot be overridden from the AI side.

Can the AI change or reprocess a payout?

No. It cannot fix, adjust, or reprocess a payout. Of the 11 tools behind the connector, 10 are read-only. The single exception saves a new report, and only after the user previews it and explicitly approves. Report creation is a separately enabled capability, so confirm it is switched on for your account.

What happens when someone leaves the company?

Deactivate them in Visdum the way you normally would. Their access stops shortly afterward, with no separate step in Claude.

Do we need to install anything or run an IT project?

No. There is nothing for IT to deploy and nothing for an end user to install. Adding the connector is configuration, and users sign in with the Visdum login they already have. It reads your existing plans, feeds, and rate tables as configured, so nothing needs restructuring.

Does it work with AI tools other than Claude?

The connector targets Claude. It is added inside Claude using a URL Visdum provides, and it authenticates each user through Visdum's own sign-in.