Release Notes

June 9, 2026

Enhanced observability — Integrated Sentry's MCP server monitoring for automatic performance tracing of all tool calls, resource reads, and protocol operations — no changes to how tools behave

Query result fix — Fixed an issue where dataset queries returning markdown-format results would show "No rows returned" despite data being available
Improved error handling — Reduced noise in error tracking by classifying expected errors (not found, validation) separately from real failures, and added resilience for partial GraphQL responses from legacy records

Memory leak fix — Resolved two memory leaks (duplicate resource registry entries and unclosed MCP sessions) that caused gradual memory growth and eventual service restarts after ~3 days
Heap observability — Added memory monitoring metrics and automated status page alerting for memory pressure

Major update with composite tools, richer responses, and interactive views.

Composite tools — 10 new high-level tools that combine multiple operations into single steps: beem_workspace_overview, beem_dataset_details, beem_explore_schema, beem_search, beem_pipeline_health, beem_create_and_deploy_dataset, beem_create_and_deploy_dataset_test, beem_deploy_and_wait, beem_build_dashboard, and beem_await_operation
Edit dataset schedule — New beem_edit_dataset_schedule tool lets you add, change, or remove a dataset's automatic refresh schedule in one step — no need to manually redeploy
Deep links in every response — Tool results now include clickable "Open in BEEM" links for datasets, dashboards, flows, sources, and other entities, taking you straight to the right page in the web app
Interactive cards — Supporting clients (Claude Desktop, claude.ai, Cursor, VS Code Copilot) now render rich HTML cards with status indicators, row counts, and action buttons alongside tool results
Server-side polling — Long-running operations (deploys, queries) now poll server-side with a circuit breaker, eliminating repeated client round-trips
Smarter error messages — Tool errors now include actionable hints and suggested next steps instead of raw error codes
Security hardening — Added rate limiting, persistent client registrations across restarts, and cache TTL enforcement

Landing page — Visiting the server URL in a browser now shows a branded page with setup instructions and links to documentation

Reliability improvements, a new tool, and longer sessions.

Source configuration tool — New beem_get_organization_source_config tool lets you inspect your organization's source configuration (e.g. Tempo, Procore), with optional filtering by source and table key
Improved dataset test reliability — Dataset test creation now self-heals when the backend returns an incorrect resource identifier, matching existing behavior for datasets
Auth improvement — Sessions now stay active as long as you use BEEM at least once every 7 days
Reliability & observability — Added error tracking, usage analytics, and automated status page incident management to improve service quality and transparency
Security fixes

Initial release of the BEEM MCP Server — connect your AI assistant to the BEEM data platform.

Explore your data — Browse workspaces, folders, datasets, and data sources using natural language
Query datasets — Retrieve rows from any dataset or preview SQL results before committing changes
Build SQL views — Create datasets, write SQL queries, and deploy them as live views with scheduling
Data quality testing — Create and run tests against your datasets to validate data integrity
Monitor pipelines — Check pipeline execution history, identify failures, and review DPU usage
Manage dashboards — Create dashboards, add widgets, and build visualizations from your datasets
Track data flows — View how datasets connect to destinations through flows
Works everywhere — Available on Claude.ai, Claude Desktop, Claude Code, GitHub Copilot, ChatGPT, and any MCP-compatible client