Cursor

Auto-review Run Mode lets agents run longer with fewer prompts, routing shell, MCP, and fetch calls through an allow-list or classifier before asking.¹

Agent runtime extended with smarter execution controls.

• Auto-review applies to Shell, MCP, and Fetch tool calls; allowlisted calls run immediately, sandboxable ones run in the sandbox, others go to a classifier subagent²
• Plans now execute in parallel via "Build in Parallel," running independent tasks simultaneously using async subagents³
• /multitask dispatches async subagents to tackle larger tasks in chunks rather than queuing them serially⁴

Canvases became shareable team artifacts. Canvases — interactive dashboards and custom interfaces created by agents — can now be shared as live browser links. Team members get read-only access via the Cursor Dashboard; available on Pro, Teams, and Enterprise plans.⁵

Cloud agent integrations reached Jira and Microsoft Teams.

• Mention @Cursor in a Jira ticket or Teams channel to trigger a cloud agent scoped to that task⁶
• Automations now support multiple attached repos or no repo at all, with five new templates; available in the Agents Window⁷
• Development environments ship as code: Dockerfile-based config, build secrets, and dependency installation for repeatable agent VMs⁸

Composer 2.5 and the Cursor SDK broadened the model and API surface.

• Composer 2.5 improves sustained long-task performance and instruction-following⁹
• The @cursor/sdk TypeScript package exposes the same agent runtime programmatically for custom pipelines¹⁰

Security and enterprise admin controls tightened.

• Security Reviewer checks every PR for auth regressions, privacy risks, and prompt injection; Vulnerability Scanner runs scheduled codebase scans (both in beta on Teams/Enterprise)¹¹
• Bugbot gained self-improving learned rules, MCP support, and configurable effort levels¹²
• Enterprise model allow/blocklists now operate at the provider and configuration level; existing blocklists must migrate by June 1¹³