Bugbot Learned Rules and MCP Support

Source: Cursor Date: 2026-04-08 URL: https://cursor.com/changelog/04-08-26

Summary

Bugbot gains Learned Rules: it analyzes PR feedback and self-generates candidate rules that promote or disable automatically based on usefulness. MCP server support for Bugbot arrives for Teams/Enterprise (enhanced code review context). Fix All action applies multiple Bugbot fixes simultaneously. Resolution rate reported at 78%, described as “highest to date.” Autofix now scoped to substantial findings only to reduce noise.

Implications

Agent-IDE feature race. Learned Rules is the first self-improving agent loop in Cursor’s product — Bugbot watches PR feedback and updates its own review behavior. This is a meaningful step toward autonomous code quality maintenance: the agent gets better at your codebase without manual rule authoring. No competitor has shipped self-improving PR review at this resolution rate.

MCP/protocol adoption. MCP support for Bugbot (not just the IDE) means MCP is expanding from interactive agent context to background process context. A Bugbot connected to a Datadog MCP can review PRs with production observability data — that’s a qualitatively different review capability than static analysis.

Watch: Whether Learned Rules are exportable/auditable (users need to trust what the agent learned) and whether the 78% resolution rate holds as Bugbot gains access to more external context via MCP.