Tool-Level Observability Hits Claude MCP Connectors

On June 8, 2026, Anthropic released new observability features and in-app directory submission tools for developers building on the Claude platform. The update targets creators of servers using the Model Context Protocol (MCP), providing a dedicated performance dashboard and a streamlined workflow for submitting custom integrations. The release is currently in public beta.

Connector Telemetry and Debugging

Developers with published connectors now have access to a dashboard that tracks usage and operational health. The interface surfaces adoption metrics, including active users, total tool calls, and directory ranking over time. This data provides a baseline for evaluating how updates impact user engagement.

Metric Category	Available Telemetry
Adoption	Active users, total tool calls, directory rank
Health	Composite health score, latency, overall error rate
Debugging	Per-tool error breakdowns, specific function failures
Surface Tracking	Usage splits across Claude, Claude Code, and Claude Cowork

The dashboard introduces a composite health score alongside raw metrics for error rates and latency. For granular troubleshooting, the interface breaks down errors at the individual tool level. This allows developers to isolate specific functions within a connector that consistently fail when invoked by Claude.

Usage data is categorized by product surface. You can view engagement distribution across the standard Claude interface, the CLI-first Claude Code, and the browser-based Claude Cowork agent. This segmentation helps developers optimize their tool schemas for the specific interfaces where their connectors see the highest traffic.

Directory Integration and Access

The update simplifies the distribution pipeline for new MCP servers. Developers can now submit their servers to the Claude Connectors Directory directly within the Claude interface. The directory currently hosts over 300 third-party connectors used across the ecosystem by millions of users.

Accessing the new observability features requires specific account tiers and permissions. The tools are restricted to Team and Enterprise accounts. Users must hold Admin or Owner roles, or possess a custom role with Libraries permissions on Enterprise plans. The dashboard and submission tools are located under the Directory tab in Organization settings.

Enterprise Security and Ecosystem Alignment

The introduction of developer-facing LLM observability aligns with a broader push for operational visibility across the Claude platform. Robust telemetry is critical as usage shifts from conversational chats to automated workflows. On the same day, Datadog announced a Cloud SIEM integration for the Claude Compliance API. This allows enterprise security teams to monitor Claude activity alongside their standard infrastructure signals, ensuring that automated actions remain within compliance boundaries.

These platform updates follow the May 28 launch of Claude Opus 4.8. The latest Opus model introduced enhanced agentic reasoning and advanced tool-calling capabilities. Models with high reasoning capabilities generate more complex tool sequences, which directly rely on stable, low-latency MCP connectors to interact with external databases and software systems like Harness. The new observability tools provide the data necessary to ensure those custom connectors can handle increased loads without degrading the user experience.

If you maintain a published MCP server, review your health score in the new dashboard to identify high-latency endpoints or failing tool calls before planning your next version release.