Your agent can now tell drops you configured apart from data you're losing, find the right attribute in projects with many fields, and spend less time verifying project-scoped connections.

These updates reached the hosted Sentry MCP server at mcp.sentry.dev on October 21 and 22, 2026.

Intentional vs. unintentional drops

Each find_dropped_events bucket now has a tags list. intentional marks drops caused by your own configuration, such as sampling, inbound filters, or before_send. unintentional marks drops you didn't ask for, such as quota, abuse, or invalid data. By Saraj Singh Manes in #1560.

Better field choices in search

When search_traces, search_logs, or search_metrics write a query from your plain-language request, they now see Sentry's description of each attribute and can page past the first 50 fields. Custom attributes that used to be cut off are reachable, and fields are chosen by what they mean, not just by name. By Nikki Kapadia in #1505 and #1518.

Faster project-scoped connections

If your connection is scoped to a single project, the project check that runs before requests now fetches less data. If that check times out, the server skips it for 60 seconds instead of making each request wait up to five seconds again. By Miguel Betegón in #1551.

Fixes

• Search tools no longer fail when a time series returns text values, such as max(timestamp) or latest_event(). Text series keep their exact values and skip peak and total summaries. With Miguel Betegón in #1424.
• create_project with a repository no longer fails when the organization has repositories that aren't linked to an external provider (#1297).
• Actions taken through Claude's hosted apps now show up in issue activity as Claude instead of Claude Desktop (#1516).

Are you using the stdio MCP server?

There was no new @sentry/mcp-server release in this period. The latest version is still 0.42.0, so these changes are only on mcp.sentry.dev until the next npm release.