CData boosts Connect AI with secure MCP agent tools

CData has expanded its Connect AI platform with new tools for building agent functions, plus identity and access controls aimed at production AI systems that interact with business data.

The product is built around the Model Context Protocol (MCP), which has gained traction as a way for AI agents to connect to business applications and data services. CData positions Connect AI as a managed MCP platform, with the latest updates focused on connectivity, context management, and governance.

Many organisations have increased spending on AI but still struggle to move projects from pilots to live use. CData argues that data access and controls account for much of that gap, particularly when agents need to read and write live data across multiple systems.

CData cited findings from its State of AI Data Connectivity Report: only 6% of organisations are satisfied with their current data infrastructure for AI. More than half still rely on custom-built integrations, and 71% of AI teams spend over a quarter of implementation time on data integration.

"AI agents are only as effective as the tools they can access and the data behind them, and only as safe as the controls governing both," said Amit Sharma, CEO and Founder of CData. "This release gives teams the ability to build use-case-specific agent tools with the right business context, deploy them securely, and enforce granular controls over which data agents can access, which actions they can take, and under what identity. That's what's been missing-not better models, but the connectivity, context, and control that make agents trustworthy enough to run in the enterprise."

Broader connectivity

A new Connect Gateway extends access to data sources behind corporate firewalls. CData lists support for SAP, SQL Server, and PostgreSQL. Connect AI also provides read and write access to more than 350 systems without replication or data movement.

The emphasis on live connections reflects a shift from analysis toward agents that trigger actions inside applications. That shift increases the need for access controls, auditing, and predictable behaviour-especially when workflows span systems such as CRM, ERP, and data warehouses.

Agent tools

CData also introduced new categories of tools that agents can use through MCP. The goal is to limit the actions exposed to a model and keep context focused on what a given agent needs.

Universal Tools provide a standard set of operations that work across connected systems. Source Tools map to defined operations within each system. Custom Tools let organisations define operations for specific workflows, including explicit data access limits and pre-optimised queries.

Two additional constructs sit alongside these tool types. Workspaces define which datasets, schemas, or views an agent can access. Toolkits define which Universal, Source, or Custom Tools are available. Each Workspace-and-Toolkit combination can be deployed as a dedicated MCP server, limiting what an agent can see and do.

Identity controls

On governance, Connect AI uses per-user authentication, applies source-system permissions at runtime, and maintains audit trails. CData also added support for SCIM 2.0 for automated identity provisioning and deprovisioning. It introduced Custom OAuth Applications as well, allowing organisations to use first-party credentials to meet internal security and compliance requirements.

Benchmark claims

CData published benchmark results comparing five MCP providers across 378 prompts in four source categories: CRM, project management, data warehouse, and ERP. Responses were scored against pre-established ground truth, with no partial credit.

According to CData, Connect AI achieved 98.5% accuracy, with 67 of 68 correct responses. Other providers ranged from 65% to 75%.

CData said errors clustered around relative date logic, multi-filter queries, semantic interpretation of business terms, and write operations. It also argued that compounding errors reduce success rates for multi-step workflows. The company attributed its results to a relational abstraction layer with semantic intelligence, rather than direct translation of natural language into API calls.

Customer comments

AnySoft, which describes itself as an agentic coding tool provider, said it uses CData to access multiple customer systems.

"AnySoft is an agentic coding tool that builds software that fits your business by ingesting, enriching, and unifying an organization's live business data-giving AI the context it needs to power everything from CRM to marketing automation," said AnySoft CEO Alex Noe. "That means connectivity to every system our customers run on isn't a nice-to-have, it's foundational. CData Connect AI Embed gives them production-grade access to 350+ data sources with the control and compliance built into the data layer."

Foodtastic also pointed to context management as a differentiator when querying enterprise systems through AI assistants.

"Connect AI has been a game changer for how we retrieve information from our enterprise systems using AI, and the reason it works so well is how it manages context," said Paul Kantorovich, Manager of FP&A and AI Strategy at Foodtastic. "It gives Claude intermediary steps to scope a request before retrieving data, unlike alternatives that load full tables and overwhelm the context window."

CData said it is demonstrating the updates at the Gartner Data & Analytics Summit, and that Chief Product Officer Ken Yagen is scheduled to speak alongside Microsoft Partner Director of Product Management James Oleinik in a session on AI agents and production deployments.