July 2025 Summaries

Model Context Protocol (MCP) makes it easier for AI systems to access live enterprise data and perform actions, but direct database connections can create risks such as destructive queries, exposure of sensitive information, excessive costs, and insufficient auditing. The proposed approach is to place APIs between AI platforms and internal data systems, creating a curated interface that limits the data, operations, credentials, and permissions available to AI while enforcing governance controls. CData positions its API Server and MCP Server as a solution for exposing real-time database, warehouse, and structured-system data through configurable API endpoints that inherit existing access permissions and support role- and object-level controls, rate limiting, IP restrictions, and centralized logging. Organizations can deploy the tools locally, define endpoints with filtering, authorization rules, and clearer field names, then connect AI frameworks through an MCP middleware layer without necessarily making APIs internet-facing. The offering is presented as a way to combine AI-driven insights with current data while maintaining security, compliance, and operational control, with a free trial and setup guide available.

CData has introduced drivers and connectors for Adobe Target that provide standards-based, SQL-accessible integration with more than 200 BI, analytics, reporting, AI, database, and custom application tools. Adobe Target, part of Adobe Experience Cloud, supports personalization and optimization through A/B and multivariate testing, audience targeting, AI-driven experiences, and performance tracking. The CData products use the Adobe Target API to provide live access to data including activities, audiences, experiences, locations, mboxes, metrics, offers, and properties, with OAuth authentication and SQL-92 query support. Available options include JDBC, ODBC, ADO.NET, native Excel, Tableau, and Power BI connectors, plus Python and PowerShell tools. A Power BI Desktop example describes installing the connector, configuring an OAuth-enabled ODBC DSN, selecting the CData Adobe Target source in Power BI, and importing data to create visual dashboards.

Model Context Protocol (MCP) standardizes connections between AI agents, large language models, and external business tools, but its rapid adoption has outpaced security practices, exposing organizations to implementation and protocol-level risks. Research cited in the article found frequent command injection, unrestricted URL fetching, file exposure, publicly accessible unauthenticated servers, and high-severity remote code execution vulnerabilities, while MCP’s flexible security design places substantial responsibility on developers. AI agents further complicate authentication and authorization because prompt injection, non-deterministic behavior, credential handling, and cross-system token mapping can produce excessive or unintended access. The article contrasts insecure experimental marketplace projects with enterprise requirements for trusted vendors, audits, support, incident response, and integration with existing security infrastructure. It highlights approaches from 1Password, which prevents agents from directly receiving raw credentials, and Epic AI, which applies just-in-time authentication and separates public from sensitive capabilities. Recommended protections include OAuth 2.1 with PKCE, least-privilege permissions, scoped and rotating tokens, code audits, input validation, allowlists, comprehensive logging, hardened and isolated infrastructure, regular patching, and tested incident-response plans.

Open-source local LLMs such as Meta’s Llama are becoming more appealing to organizations because they offer greater control, customization, privacy, compliance support, and freedom from vendor lock-in compared with cloud-hosted models. Their main limitation is that their knowledge can be static and disconnected from the current enterprise data needed for accurate, context-aware responses. Model Context Protocol (MCP), an open-source standard, addresses this limitation by securely connecting LLMs to live business systems, interpreting prompts, running parameterized queries, and supplying structured results to the model when needed. CData MCP Servers are presented as a way to connect locally deployed models with governed data from sources such as Salesforce, including through an LM Studio setup that involves configuring an MCP server, downloading a model, and chatting with connected data. This combination can support data-aware copilots, chatbots, analyst workflows, executive reporting, and customer service automation while reducing data duplication, custom integration work, and the need to construct new data pipelines.

CData’s sixth Vibe Querying episode demonstrates how its beta Asana MCP Server connects AI clients such as Claude and Gemini to live Asana data, allowing project managers to explore, analyze, and act on project information through natural-language questions. In a demonstration with marketing project manager Luc Leblanc, Claude identified overdue and unassigned tasks across several projects, deleted obsolete 2022 tasks, investigated a Forrester report containing 20 unassigned subtasks, created a priority- and deadline-based personal task list, generated a weekly status report, and summarized completed milestones and upcoming editorial-calendar deadlines. The post argues that this approach reduces the manual work of navigating projects, applying filters, compiling reports, and finding cleanup opportunities, while CData’s SQL-based interface can manage complex relationships among tasks, subtasks, projects, and users and support both data retrieval and write actions. CData presents conversational “vibe querying” as a way for project managers to shift attention from administrative tracking toward resolving bottlenecks, coordinating teams, and making strategic decisions.

CData’s Summer 2025 Product Showcase emphasized that successful enterprise agentic AI depends on secure, governed, actionable access to connected data rather than passive query capabilities alone. Its new Model Context Protocol Servers, based on Anthropic’s MCP standard, are designed to let LLMs securely read and write across more than 270 enterprise data sources through structured queries and stored procedures, with Claude support in beta and a SaaS-oriented remote server planned for CData Connect AI. The company also announced enhancements to CData Sync, including Microsoft Mirroring for Delta replication in OneLake, additional reverse ETL destinations, and natural-language job orchestration, alongside planned Iceberg, dependency, and MCP capabilities. CData Arc added Python scripting, improved REST connectivity, and Peppol e-invoicing support, while CData API Server offers rapid API generation with controls such as rate limiting, access management, and logging. Audience feedback identified ETL and API integration as leading data use cases and safe, governed data access as the primary challenge for AI agents, reinforcing CData’s focus on infrastructure for reliable enterprise AI deployment.

CData has introduced drivers and connectors for Jira Assets, the asset and configuration management component of Jira Service Management, to enable access to asset data from BI tools, databases, custom applications, and automation environments. The offerings include JDBC, ODBC, ADO.NET, Excel, Tableau, Power BI, Python, and PowerShell integrations, with features such as live data access, SQL-92 querying, API-token authentication, and read/write capabilities. The connectors present Jira Assets data in a relational model, exposing objects, schemas, object types, attributes, statuses, and import sources as tables or views. An example workflow for Power BI involves installing the connector, configuring an ODBC data source with Jira credentials, selecting Import or DirectQuery mode, and building reports that track metrics such as asset distribution, ownership, relationships, and lifecycle statuses.

AI-generated data drivers use large language models or templates to rapidly create basic connectivity code from API specifications and schemas, offering faster prototyping, lower initial development effort, and reusable integration patterns. The passage argues that these drivers may lack advanced capabilities, testing, performance optimization, maintainability, detailed error handling, and enterprise security needed for production-scale workloads. In contrast, CData’s purpose-built drivers are presented as domain-expert-engineered components that provide SQL-based access, system-specific optimizations, lifecycle maintenance, authentication support, metadata discovery, and compatibility with common data tools. A comparison between an AI-generated Jira plugin and CData’s Jira driver is used to illustrate differences in authentication methods, API coverage, CRUD support, relational data modeling, performance features, logging, and support for schema changes. The central argument is that while generated drivers can reduce short-term development time, organizations should assess long-term reliability, maintenance costs, security, and scalability when making data-connectivity decisions.

CData presents “vibe querying” as a conversational analytics approach in which AI agents such as Claude Desktop access live enterprise data through Model Context Protocol servers and a standardized SQL-92 interface rather than custom API integrations. Using more than 350 connectors, the proposed system can query data in place across sources including CRM, marketing, support, analytics, and inventory platforms, allowing users to ask follow-up business questions about sales pipelines, marketing performance, support trends, operations, and churn without manually preparing reports or moving data into a consolidated store. The company argues that SQL provides a more consistent interface for AI than source-specific APIs, reducing setup and maintenance while supporting features such as query pushdown, parallel execution, and bulk operations. It also states that access remains governed by each user’s existing source-system credentials and permissions. The approach is positioned as a complement to existing BI tools that could shift data teams from producing static reports toward facilitating real-time exploration, with CData MCP Servers available in beta for Claude Desktop and additional AI platforms planned.

Google Sheets’ experimental =AI() formula enables generative AI prompts directly in spreadsheet cells, while CData Connect Spreadsheets extends this capability by connecting Sheets to more than 270 live enterprise data sources, including CRM, accounting, e-commerce, analytics, databases, and warehouses. Users can enable Workspace Labs features, connect their systems through CData, apply AI prompts to synchronized data, and refresh both source data and generated outputs as needed without coding or manual exports. Suggested applications include summarizing general ledger records and identifying accounting anomalies, generating product descriptions and marketing copy from SKU data, and creating personalized sales outreach, deal summaries, and customer-review responses from CRM or support information. The service is presented as a way to combine current business data with spreadsheet-based AI automation at scale, with a free trial available.

CData Connect AI has expanded its Azure Data Factory integration with SQL Server Integration Services component connectivity, adding a third connection option alongside OData endpoints and its Virtual SQL Server API. The SSIS-based approach is intended to improve high-volume data transfer performance and support complex ETL tasks, including row-level transformations, fuzzy matching, lookups, cleansing, and custom C# or VB.NET script tasks. It also aims to simplify connections to varied data sources and support existing legacy SSIS packages, while scaling for increasing data volumes. Users can begin by creating a Connect AI trial, adding source connections and optional datasets, authenticating ADF through OAuth, and then accessing configured data and datasets within ADF.

Snowflake Openflow is a newly released, Apache NiFi-based ingestion service designed to support real-time structured and unstructured data pipelines into Snowflake, with native Cortex AI integration and customization opportunities, but it currently has a relatively small connector catalog, requires hands-on flow development, is limited to AWS bring-your-own-cloud deployments, and uses volume- and compute-based pricing. CData Sync is presented as a mature alternative focused on no-code enterprise replication across cloud, on-premises, and hybrid environments, offering more than 250 connectors, reverse ETL, configurable transformations, and deployment options across hosted, AWS, Azure, and on-premises systems. The comparison positions Openflow as most appropriate for organizations operating primarily within the Snowflake ecosystem and seeking AI-oriented, real-time ingestion, while positioning CData Sync for teams requiring broad source and destination compatibility, predictable connection-based pricing, and flexibility across diverse existing architectures.