cursor 3.11: side chats, team mcp distribution, and cloud agent hooks

Cursor v3.11, released July 10, 2026, makes four changes that are individually useful and collectively significant: side chats for parallel agent conversations, searchable transcripts for past sessions, team-wide MCP server distribution from an admin dashboard, and programmable lifecycle hooks for cloud agent runs. The through-line is Cursor moving from a single coding assistant toward something closer to a managed agent workspace.

side chats: parallel investigations without context-switching

The most visible addition is side chats. You can open one with /side, /btw, or the plus button in the chat panel. Each side chat inherits context from the main conversation, runs independently, and stays durable — you can come back to it later and reference it back into the main thread with an @-mention.

By default, side chats focus on reading, searching, and answering rather than editing. This is deliberate: the intended use case is investigating a question while your primary agent session continues working. You want the side thread to gather information, not start making changes that could conflict with what the main thread is doing.

The practical payoff: you can run a primary implementation task, open a side chat to explore an alternative approach or pull in documentation, and then bring the side chat's findings back into the main session when you need them. No context-switching, no losing your place.

agent transcript search

Cursor 3.11 ships a local search index over all agent conversations. Cmd+K (or Ctrl+K on Windows) in the Agents Window opens cross-session search against actual message content. Cmd+F within a single conversation gives match-jumping and a match counter.

The index builds locally, so agent transcript content does not route through Cursor's servers. If you work on anything sensitive — API design, internal tooling, commercial code — this matters. The content stays on your machine.

The search is useful in proportion to how long you've been using Cursor. Early on there's not much to find. A few months in, when you're trying to recover an approach you worked through three weeks ago in a session you can't reconstruct from memory, it starts earning its place.

team mcp distribution

Admins on Team and Enterprise plans can now configure MCP servers once and distribute them to every team member. The servers appear in a team marketplace that members can install locally, and cloud agent runs pick up the same configuration automatically — no per-developer setup required.

Before this, getting consistent MCP tool access across a team meant everyone managing their own configuration independently. That works at small scale and breaks down quickly as teams grow or MCP server config gets more complex. Admin distribution solves the coordination problem without requiring developers to think about it.

The feature requires a Team or Enterprise plan. The per-member marketplace install flow is available on Team; enterprise-managed distribution (where servers are configured centrally and pushed rather than pulled) requires Enterprise.

cloud agent hooks

Five new lifecycle hook events are available for cloud agents via a repo-level .cursor/hooks.json file. Hooks fire on prompts, responses, agent thoughts, subagent spawns, compaction events, stops, and turn completions.

The hooks enable a category of automation that wasn't possible before: inspecting what the agent is about to do before it does it, logging agent activity to an external system, applying policies that modify agent behavior at runtime, or building self-correcting workflows where a hook can catch an undesirable agent state and redirect it.

Hooks require a .cursor/hooks.jsonfile at the repo root. There's no plan restriction — any user can define them on any repo they control. Enterprise-managed hooks (centrally distributed, not per-repo) require an Enterprise plan.

the broader pattern

Cursor's previous capability gap versus Claude Code was most visible in multi-task workflows. Claude Code's parallel subagent model let you run multiple workstreams from a single session; Cursor required you to context-switch between separate windows or lose the thread of one task to start another. Side chats are a direct answer to that gap — not identical to Claude Code's subagent model, but covering most of the same use cases.

Team MCP distribution and cloud agent hooks are in a different category: they're about making Cursor manageable at the team and org level rather than just the individual developer level. That's where the product needs to go if it's competing with GitHub Copilot for enterprise accounts — Copilot's admin tooling and GitHub Actions integration have been the clearest enterprise advantages Cursor lacked. Version 3.11 narrows that gap.