November 2025 Summaries

CData has enhanced its managed MCP platform, Connect AI, with source-specific instructions for hundreds of enterprise connectors, aiming to help AI applications understand schemas, metadata, entity relationships, dynamic objects, and common operations before querying data. The enhancement addresses common large language model limitations, including difficulty navigating large catalogs, discovering relationships, applying exact filters, avoiding invalid queries, and safely performing write operations, which can otherwise cause failed requests, excessive API use, high token consumption, and costly full-dataset scans. Drawing on more than a decade of connector development, the instructions provide mapped core entities, dynamic-schema discovery steps, progressive query guidance, stored-procedure examples, and relationship models tailored to systems such as Salesforce, NetSuite, and ServiceNow. CData says the update improves accuracy, response speed, cost predictability, and time to value through semantic metadata, query push-down, derived views, curated multi-source data collections, custom tools, and document access within AI workflows.

Real-time enterprise connectivity is presented as an alternative to batch, streaming, and vector-database pipelines for supplying LLMs with current operational data. Rather than replicating information into separate stores, permission-aware connectors and Model Context Protocol servers can let models query live systems such as ERP, CRM, HRIS, SQL databases, and SaaS applications at request time, improving freshness, reducing synchronization drift, lowering maintenance, and limiting governance risks from duplicated sensitive data. The approach supports on-demand transformations, role-based access controls, filtering or masking, audit logging, and monitoring of requests, permissions, availability, and response times. It is described as particularly useful for regulated industries and applications including financial risk analysis, healthcare workflows, manufacturing inventory and equipment monitoring, and customer service, while integrating with AI frameworks such as LangChain, CrewAI, Copilot Studio, and function-calling tools. CData positions its Connect AI managed MCP service as a unified platform for this architecture, arguing that direct live access can produce more grounded LLM responses than pipeline-based retrieval systems.

Model Context Protocol (MCP) is presented as a standardized framework for connecting large language models, AI agents, and automation tools to live enterprise data while preserving business context, security controls, and auditability. CData Connect AI is described as a managed MCP platform that provides governed, real-time access to more than 300 data sources without replicating data, inheriting authentication, permissions, and security policies from underlying systems. The platform aims to address limitations of traditional APIs and batch ETL, including delayed data, lost semantics, fragmented integrations, and inconsistent governance, through centralized connectivity, identity-based logging, role-based access, and secure tool invocation. It supports legacy, cloud, on-premises, and hybrid environments through standard protocols such as SQL, ODBC, and JDBC, and is positioned for uses ranging from embedded product features to sales, finance, and IT automation. The material also emphasizes compliance alignment, risk mitigation for threats such as permission leakage and data exfiltration, and future support for multi-agent workflows, lifecycle management, semantic intelligence, and evolving AI governance requirements.

AI adoption is compared to the early electrification of factories, where productivity gains were delayed because organizations applied new technology to workflows designed for older systems rather than redesigning operations around its capabilities. Although many companies have modernized through cloud platforms, APIs, data lakes, ETL pipelines, and governance, these architectures are often not optimized for AI’s much greater requirements for context, reasoning, data variety, volume, and real-time information. The passage argues that organizations achieving stronger AI results are treating AI as an active participant in workflows that can retrieve context, support decisions, validate outputs, identify inconsistencies, and initiate actions across systems, rather than as a reporting layer added to business-intelligence-era infrastructure. It advocates selectively reworking the architectural components most stressed by AI, potentially including refined pipelines, live-data layers, and semantic discovery, and promotes Mark Palmer’s e-book, Six Moves to Rewire Software for the AI Age, as a blueprint for AI-native data architecture.

Microsoft and CData announced a collaboration at Microsoft Ignite 2025 to integrate CData’s managed MCP platform, Connect AI, with Copilot Studio, Azure AI Foundry, and Agent 365. The partnership focuses on enterprise AI agent requirements for connectivity, context, and control, using more than 350 prebuilt connectors, semantic metadata about source systems, and identity-based security controls such as OAuth/SSO, role-based access, least-privilege permissions, and audit trails. Within Copilot Studio, the integration is intended to extend agent access to enterprise applications and databases including SAP, Salesforce, Oracle, ServiceNow, Snowflake, and legacy systems, while Azure AI Foundry users can build applications that analyze data across multiple sources. Agent 365 is positioned as the governance layer for agent behavior, complemented by Connect AI’s controlled source-system access and monitoring capabilities. CData argues that combining broad data access, system-level context, and security oversight can help organizations move AI agents from experimental demonstrations to production tools for cross-system analysis, workflow automation, and business insights.

CData Sync has added change data capture support for IBM DB2 iSeries (AS400), extending its existing CDC capabilities for DB2 on z/OS and Linux, Unix, and Windows to cover the full IBM DB2 family. The hybrid replication platform is positioned as a way for organizations to move data securely and near real time from long-standing on-premises IBM systems into cloud and analytics destinations such as Databricks, Snowflake, Microsoft Fabric, and SQL Server without disruptive migrations. Its iSeries connector uses IBM i’s native journal records to capture inserts, updates, and deletes, avoiding triggers and full-table scans while aiming to preserve transactional accuracy, limit production impact, and maintain continuity through restarts or outages. CData argues that this support helps enterprises retain the operational value and historical context of IBM databases used in sectors such as banking, manufacturing, logistics, government, and telecommunications while making their data more accessible for reporting, analytics, AI, and machine learning initiatives.

Real-time data pipelines give LLM applications continuous access to current enterprise data, improving response relevance, accuracy, and adaptability compared with traditional batch ETL processes. Effective pipelines combine multi-source ingestion, data cleansing and validation, text embedding and vector storage, workflow orchestration, and continuous monitoring to support reliable retrieval-augmented generation and other AI uses. Organizations must address fragmented systems, latency, scalability, and compliance through governed connectivity, source-level permissions, federated queries, semantic layers, and data lineage controls. Technologies such as Kafka, dbt, Pinecone, LangChain, Airflow, Kubernetes, Arize AI, and Weights & Biases support different pipeline stages, while CData Sync and CData Connect AI are presented as platforms that provide low-latency, permission-aware connectivity and replication across more than 350 data sources. These capabilities can enable applications including current financial reporting, consolidated customer insights, and LLM responses grounded in live business knowledge.

AI agents operating on behalf of users should be treated as applications rather than independent non-human identities, using established OAuth, SSO, and delegated authorization patterns instead of new IAM accounts and credentials. Assigning agents separate identities can produce excessive privileges, account sprawl, weak attribution, and greater risk because LLM-driven agents are less predictable than conventional software. The proposed approach enforces permissions when an agent executes a specific action on a resource for a particular user, with access governed by scoped connections and auditable APIs. Connect AI applies this model by using credentials linked to users, preventing agents and models from handling access tokens, and limiting actions according to data-source and platform-level permissions. The company argues that separating user identity, model reasoning, and application execution allows developers to avoid embedding identity logic, gives security teams centralized control, and enables scalable, traceable agent deployments without creating new IAM constructs.

CData’s 2025 Q4 Driver and Connector release adds support for Klaviyo, Databricks Lakebase, and Oracle Eloqua Reporting, expanding SQL-based access to marketing, database, and reporting data for analytics, integration, and AI workflows. The release also modernizes Neo4j connectivity for Neo4j 5.x and AuraDB, improves WordPress compatibility, and enhances MCP server capabilities for Google Drive and Sheets, including direct Google Docs content access and expanded file-management features. SQL Gateway now supports PostgreSQL endpoints, Microsoft Graph drivers gain OAuth scope configuration for sovereign cloud environments, and Databricks, Sage Intacct, Shopify, BigCommerce, Salesforce Account Engage, and QuickBooks Online receive significant updates, including API migrations, authentication changes, and expanded data operations. Additional improvements span platforms such as HubSpot, Snowflake, Google BigQuery, Amazon Athena, LinkedIn, Mailchimp, SharePoint, SurveyMonkey, and SAP BusinessObjects BI, reflecting hundreds of portfolio-wide updates and compatibility enhancements.

As enterprises expand use of large language models such as ChatGPT, Microsoft Copilot, and Google Gemini, secure managed access to live enterprise data is presented as essential for preventing data exposure, maintaining compliance, and preserving operational oversight. Effective LLM governance includes identity-based authentication, role-based access controls, encryption, audit logs, rate limiting, continuous monitoring, incident response, and testing against threats such as prompt injection, model inversion, and excessive agency. Organizations are advised to align LLM security with business priorities including data protection, cost control, and stakeholder trust, while mapping controls to regulations such as GDPR, HIPAA, and CCPA and using frameworks including OWASP, MITRE ATLAS, and NIST. The text contrasts real-time, governed data connectivity with data-replicating ETL approaches, arguing that live access can reduce latency, duplication, and compliance complexity. It highlights CData Connect AI as a no-code platform using the Model Context Protocol to connect multiple AI models to more than 270 enterprise data sources while retaining source-system permissions, supporting cloud and hybrid environments, and providing security features such as OAuth, SSO, MFA, RBAC, logging, and encryption.

Real-time SAP-to-Databricks integration can improve decision-making, analytics, automation, and AI by making operational data quickly available in Databricks, but it requires careful planning around latency, schema changes, security, and data quality. The recommended approach begins by defining measurable business outcomes and KPIs such as freshness, processing latency, completeness, and integrity, then selecting a batch, event-driven, or hybrid architecture based on operational needs. The guide positions CData Sync as a low-code tool for connecting SAP and Databricks, supporting secure authentication, role-based access, governance, full loads, and timestamp- or integer-based incremental replication. It advises preparing SAP data and Databricks governance through accessible, standardized source tables and Unity Catalog, choosing suitable SAP change-data-capture methods such as ODP, SLT, or database logs, and establishing a clean historical baseline before transitioning to incremental updates. Ongoing validation should reconcile record counts, monitor freshness and latency, and verify relational consistency, while dashboards, alerts, logging, and scalable pipeline design help maintain reliable performance as data volumes and use cases expand.

CData’s comparison of Claude Skills and the Model Context Protocol (MCP) argues that the two approaches serve complementary roles rather than replacing one another. MCP enables AI agents to discover and explore external systems, schemas, tools, and data sources, while Claude Skills package known, repeatable workflows into instructions and code that can execute tasks with less model context. In controlled tests using Salesforce and Zendesk data through CData Connect AI, MCP combined with Skills reduced token use by 65% for data discovery, 34% for a simple revenue query, and 58% for a cross-system join compared with MCP alone. The savings resulted from using MCP initially to identify available data and construct requests, then moving established requests into Skills that call APIs directly and return streamlined results rather than requiring the model to process extensive schema metadata and verbose JSON. The proposed workflow is therefore to use MCP for initial discovery and understanding, then create Skills for efficient, reliable execution of recurring tasks.

CData announced that its Connect AI platform is a featured Model Context Protocol launch partner in the Databricks Marketplace, intended to extend Databricks Agent Bricks with governed access to live enterprise data and operational systems. The integration is designed to let AI agents combine historical analysis in Databricks with real-time information and actions across more than 350 systems, including Salesforce, ServiceNow, and NetSuite. Connect AI provides connectors that expose system data, metadata, relationships, and business logic, while supporting both read and write operations such as updating sales opportunities, creating support tickets, and adjusting inventory forecasts. The platform also emphasizes enterprise controls including existing access permissions, row- and field-level security, audit logs, and compliance certifications, and is available through a 14-day free trial in the Databricks Marketplace.

Context engineering is presented as the practice of supplying an LLM with the right minimum amount of information, tools, and data at the appropriate time, extending beyond prompt engineering to include system instructions, memory, external data, and tool access. As RAG and the Model Context Protocol make live enterprise data more accessible to LLMs, managing context has become important for controlling costs, improving reliability, and reducing hallucinations. CData positions its Connect AI platform as a context-engineering solution built on its SQL-based data connectivity foundation, offering access to more than 350 enterprise sources. Its approach aims to limit tool selection complexity by using eight tools regardless of connected systems, provide semantic understanding of standard and custom data objects, and process complex queries through source systems or the platform before returning results to the LLM. According to the text, these capabilities reduce token use and data-model exploration while leaving more of an LLM’s context window available for analysis and actions.

Production AI agents require authorization beyond user authentication, because they must act on behalf of individual users across external services while being constrained to specific actions and resources. Generic service accounts can bypass user-level permissions, while granting agents full user credentials can create risks from errors, hallucinations, or prompt injection. The text presents CData Connect AI as a managed MCP platform that addresses this gap through user-specific, connection-level permissions enforced when tools are executed, enabling controlled actions such as querying Salesforce, updating Jira, or sending messages. It provides generalized tools for accessing and acting on business systems, abstracts OAuth flows, token management, credential storage, auditing, and service integrations, and can be used with MCP-compatible agent frameworks such as LangChain to securely retrieve data and perform permitted tasks.