MCP has changed the speed of the conversation. 

Eighteen months ago, connecting an AI assistant to your accounting data meant custom integrations and complicated build cycles. Today, with Model Context Protocol standard across every major AI platform, a firm can have its AI querying financial data in minutes. 

That shift is real. It has also created a new version of an old problem. 

The question audit firms are asking now is not whether they can connect AI to their data. It is whether the data underneath is clean, standardized, and traceable enough to trust the output. 

Why MCP matters more than most firms realize 

Model Context Protocol is the open standard that lets AI agents connect to external tools and data through a single, permissioned interface. Anthropic introduced it in November 2024 and donated it to the Linux Foundation’s new Agentic AI Foundation in December 2025. It is now supported natively by Anthropic, OpenAI, Google, Microsoft, and AWS, with more than 97 million monthly SDK downloads and over 10,000 active servers as of early 2026. 

For audit and accounting firms, the practical change is significant. Before MCP, getting AI to work with client financial data meant a data export, a reformatting step, and a manual import, every engagement, every client, every time a dataset changed. 

MCP removes those steps. An AI agent can query standardized financial data directly, with every call permissioned and logged. The audit trail is part of the connection itself, not something added afterward. 

In June 2026, MCP’s Enterprise-Managed Authorization reached stable status, with Okta as the first supported identity provider and Anthropic and Microsoft among the first to adopt it. Firms can now manage AI data access through their existing identity provider, applying policy once rather than configuring permissions for every user and every server. 

For security teams at accounting firms, this is the missing piece. The governance question has moved from ‘can we control this’ to ‘how do we configure the controls we already have.’ 

Why the data layer blocks agentic AI in audit 

The tools are impressive. An AI agent can draft AR review files, run ratio analysis, and flag control breakdowns without a single data export. But every one of those capabilities depends on the same thing: the financial data the agent queries must be standardized, complete at transaction level, and consistent across every engagement. 

When it is not, the AI does not return an error. It returns a plausible-sounding wrong answer, which is worse than a clear failure. 

The problem is structural. Accounting systems do not share a common data model. A trial balance from QuickBooks and a trial balance from NetSuite do not use the same chart of accounts, field names, or structure. An AI agent querying both without a standardizing layer in between is not running analytics. It is guessing at the mapping. 

For audit work, that variability is not acceptable. Every conclusion needs to trace back to a specific transaction. Every procedure needs to be reproducible. This is not a failure of the AI tools. It is a data foundation problem, and it is why firms that standardize their data layer first are best placed to benefit from agentic workflows. 

Where Validis fits 

Validis connects directly to leading accounting systems, cloud and on-premise, and standardizes GL, AR, AP, and Trial Balance data at transaction level to a common audit data model. That standardized dataset already powers your team’s manual procedures. It is the same dataset AI agents need to do useful, checkable work. 

Validis MCP, the AI connector for standardized financial data. One sign-in, read-only, audit-logged, built for Claude, ChatGPT, Copilot, and Excel.
Read next

 

Cloud or on-premise: the data problem does not discriminate 

A common objection to AI-powered audit workflows is client coverage: if a client runs an on-premise system, the assumption is that agentic workflows do not apply. That assumption does not hold. Standardization has to happen regardless of where the source system sits. 

A firm with a mixed client book, some on QuickBooks Online, some on Sage, some on on-premise Microsoft Dynamics, needs the same standardized layer underneath its AI for all of them. That is the coverage argument that matters most while cloud and legacy systems still sit side by side at mid-market companies. 

The governance question firms are now asking 

The conversation at accounting conferences in 2026 has shifted from whether to adopt MCP to how to govern it. With Enterprise-Managed Authorization now stable, firms can manage AI data access through their existing identity provider, and every action can be reviewed against controls already in place. 

What this means in practice: MCP access does not require a separate permission structure built from scratch. It plugs into the governance infrastructure a firm already has, or is already building. 

What standardized data already delivers 

Andrew O’Donnell, Manager at Citrin Cooperman, and Stephen Panfile, Senior Manager at Withum, have both spoken publicly about what standardized data unlocks for their firms. O’Donnell: as firms explore AI tools, you need standardization to run them effectively. Panfile: saving even a few minutes on a task, repeated across a full engagement book, adds up to real WIP saved. 

Why Citrin Cooperman and Withum won’t run an audit without Validis.
Read next

 

What this means for your firm now 

You do not need a finished AI strategy to start on this. Get the data layer right first, independent of which AI assistant your team standardizes on. If your firm already uses Validis, your data is already standardized. Ask your Account Manager about Validis MCP, and how that same standardized data connects directly into the AI tools your team already uses. 

Why quality data is now the backbone of AI-powered audit workflows, by Jeff Gramlich, CPA, CITP.
Read next

 

Talk to the Validis team

Resources 

• MCP joins the Agentic AI Foundation. Model Context Protocol Blog, (9 December 2025)
• Linux Foundation announces the formation of the Agentic AI Foundation. Linux Foundation, (December 2025)
• MCP Enterprise-Managed Authorization goes stable. InfoQ, (July 2026)
• Validis: why quality data is now the backbone of AI-powered audit workflows, by Jeff Gramlich, CPA, CITP  
• CPA Practice Advisor: what is Model Context Protocol. Accounting Technology Lab Podcast, (February 2026) 
• Thomson Reuters: how agentic AI is redefining the tax and accounting profession