Fewer failed tool calls: long-running operations no longer get cut off, and errors now tell you what to use instead of just what went wrong.

• Long operations no longer time out silently — Deploys, dataset refreshes and large queries that ran longer than a minute used to be dropped with an unexplained gateway error. They now run to completion, and one that genuinely takes too long returns a clear message naming the operation and how to resume it
• SQL errors now tell you the real column and table names — Previewing a query that references a column or table that doesn't exist now lists the actual columns of that table (or the available tables), so the query can be fixed straight away instead of by trial and error
• Better guidance on failed queries — A query rejected because of the SQL itself is no longer reported as a platform fault to escalate; it now says what to correct
• Source and connector configs are discoverable — Requesting an organization's source or Terraform config with a key that doesn't exist now returns the list of keys that do exist, and keys are matched regardless of capitalization or dashes. Very large configs return their key list instead of failing outright

BEEM MCP Server, released 10 August 2026