# Datasentinel 2026.10 adds AI Connector with MCP server for PostgreSQL monitoring

> **Key takeaway:** Datasentinel's PostgreSQL monitoring platform now integrates with major AI assistants through an MCP server, enabling natural-language database troubleshooting while keeping credentials isolated.

**Published:** 2026-11-15T08:02:18.818488+00:00 
**Source:** PostgreSQL Official News 
**Category:** databases-storage 
**Canonical URL:** https://fosswire.org/news/datasentinel-202610-adds-ai-connector-with-mcp-server-for-po.html 

## Summary
Datasentinel 2026.10 introduces AI Connector, an add-on that connects the PostgreSQL monitoring platform to AI assistants including Claude, ChatGPT, Gemini, Microsoft Copilot, and GitHub Copilot. Users can query database performance in plain language and receive analysis backed by collected metrics, execution plans, and concrete recommendations. The feature combines an MCP (Model Context Protocol) server for metric access with a guiding Skill to direct analysis, and supports chat interfaces, on-demand dashboards, scheduled health checks, and scripted API calls.

The release also brings redesigned PDF reports with lower memory usage and faster generation for large fleets, a new optional datasentinel_analytics function for read-only planner statistics access requiring Agent v4.0.0 or later, automatic purging of inactive instances, unified query monitoring filters for agentless deployments, and updated third-party dependencies for security. The team continues to maintain pg_datasentinel, an open-source PostgreSQL observability extension.

## AI Connector architecture and security model
- **MCP server and Skill pairing:** AI Connector exposes Datasentinel metrics through an MCP server while a companion Skill guides the assistant's analysis workflow, separating data access from reasoning logic.
- **OAuth 2.1 authentication:** Users authenticate through Datasentinel's login page; assistants receive revocable tokens with no access to passwords or database credentials, and permissions mirror existing web interface roles.
- **Read-only, agent-mediated live checks:** The AI assistant never connects directly to databases; live troubleshooting queries run through the Datasentinel agent in read-only mode, with execution plans estimated rather than executed.
- **datasentinel_analytics function:** A new optional internal function provides read-only access to planner statistics and execution plans, requiring Agent v4.0.0 or newer for deployment.
- **Audit and revocation capabilities:** Scripts and scheduled jobs use revocable access tokens, with sign-in and token events logged for security review.

## Why it matters
The MCP-based integration pattern demonstrated here could influence how other database and infrastructure monitoring tools expose telemetry to AI assistants, potentially establishing a de facto standard for secure, read-only observability access. For PostgreSQL administrators, the natural-language interface lowers the barrier to complex performance investigation while the layered security model—no direct database access, role-based visibility, and full audit logging—addresses enterprise compliance concerns that often stall AI adoption in production environments.