Best Ways to Track What Cursor's Agent Did in 2026
The best ways to track Cursor agent changes in 2026: WaveBrain first, then Cursor checkpoints, git history, hooks and shared transcripts, compared.
Reviewed by WaveBrain Operations TeamReviewed 6 min read
The short answer
The best way to track what Cursor's agent did in 2026 is WaveBrain, a hosted activity log that Cursor connects to over MCP, so deploys, pushes, merges and other outside actions land on one searchable timeline with failures in Slack. Cursor's checkpoints, git history, hooks and shared transcripts cover file rollback, code review and custom logging.
Cursor's agent finished a refactor across 14 files, ran the test suite, then ran a deploy script you had forgotten was in package.json. The chat shows all of it, but a week later, in a different chat, you will not remember which conversation did the deploy or whether it succeeded.
Cursor gives you good tools for undoing agent edits, and Git covers committed code. Tracking what the agent actually did, especially outside the editor, takes something else. Below we rank five ways to keep track, starting with the one that answers the question fastest.
What is the best way to track what Cursor's agent did?#
WaveBrain is the best way to track what Cursor's agent did because every action with outside effects becomes a short, searchable entry that stays put after the chat is closed. Cursor's own features are built for rolling back files and sharing conversations, which is a different job.
| Method | Best for | What it records | Where you read it | Across machines and agents | Failure alerts |
|---|---|---|---|---|---|
| WaveBrain | Knowing what the agent did: deploys, pushes, merges, messages, config changes | One entry per action with title, details, project, link and status | Web timeline with search and filters | Yes, Cursor plus Claude Code, Codex, Hermes and scripts in one place | Yes, failures posted to Slack |
| Cursor checkpoints | Undoing a wrong turn in one chat | Snapshots of files the agent modified | The chat timeline in Cursor | No, stored locally | No |
| Git history | Reviewing committed code | Commits and diffs | git log, your Git host | Yes, for pushed code | No |
| Cursor hooks | Custom logging of shell, MCP and file events | Whatever your script writes | Your own file or service | Only if you build it | Only if you build it |
| Shared transcripts | Showing a teammate a conversation | The full chat, tool calls and results | A link on cursor.com | Team members with the link | No |
How we chose: what did we compare?#
We compared ways to review Cursor agent work after it happened, based on Cursor's documentation and WaveBrain's homepage as of November 13, 2026. For each method we checked whether it shows actions outside the repository, whether you can search it later, whether it works across machines and other agents, whether it alerts you on failure, and how much setup it needs. We left price out of the ranking.
Why is WaveBrain the best way to track Cursor's agent?#
WaveBrain keeps a record of outcomes rather than file states, so you can answer "did the deploy go out?" without finding the right chat. It is a remote MCP server. You add it to ~/.cursor/mcp.json with your API key (or use the "Add to Cursor" button on the WaveBrain Setup page), and the server's instructions tell the agent to call log_action right after anything with external effects, such as a deploy, a push, a published post, a sent message or a config change. Pure reads and research are skipped.
Each entry records the agent name, an action type such as deploy, push or merge, a past-tense title, details of what changed and the result, the project, a link and a status. The dashboard filters by agent, project and type and searches titles and details, and failed actions post to a Slack channel.
The biggest advantage over Cursor's built-in tools is scope. If Cursor, Claude Code, a Hermes bot and a cron script all report to the same workspace, one timeline covers them all, which matters when you track AI agents across multiple machines.
Be clear about the limits. WaveBrain is not LLM tracing, so it records no prompts, tokens or latency, and it is not a tamper-proof audit log. With Cursor, the agent logs itself through MCP, so pasting WaveBrain's short logging rules into an AGENTS.md or Cursor rules file makes it log consistently.
What do Cursor checkpoints track?#
Cursor checkpoints track file snapshots during an Agent session so you can undo a wrong turn. Cursor's Agent docs say the agent creates them automatically before significant changes, capturing the state of all modified files, and you can click any checkpoint in the chat timeline to preview and restore.
The docs are also direct about their scope. Restoring a checkpoint reverts files only and does not remove messages, and checkpoints are "stored locally and separate from Git". Cursor's advice is to use them only for undoing Agent changes and to use Git for permanent version control. Checkpoints cannot tell you about a deploy, an API call or a message the agent sent, because none of those are files in your project.
How does git history help track Cursor's changes?#
Git history is the permanent record of code the agent changed, as long as the changes get committed. Commit before each agent run, review the diff afterwards, and you have an exact, line-by-line account of what landed in the repository.
Git has two gaps for agent work. Edits the agent made and you later discarded never reach a commit, and actions outside the repository, such as running a migration against a live database or calling a third-party API through an MCP tool, leave nothing in Git at all. That is why most people pair Git with an action log.
Can Cursor hooks log what the agent does?#
Yes, Cursor hooks can log agent activity, if you are willing to write and maintain the scripts. Cursor's hooks docs describe hooks as processes that "observe, control, and extend the agent loop", defined in hooks.json at the project level or in ~/.cursor/hooks.json for your user.
Useful events for tracking include afterShellExecution, afterMCPExecution, afterFileEdit, postToolUse, postToolUseFailure, stop and sessionEnd. The docs even show an audit.sh example wired to several events. Two caveats from the same page matter. User-level hooks in ~/.cursor/hooks.json are not available in cloud agents, and you still have to decide where the log goes and how to search it.
Are Cursor shared transcripts useful for tracking?#
Shared transcripts are useful for showing someone exactly how a conversation went, but they are a sharing feature rather than a log. Cursor's help center says a shared transcript includes the full conversation history, code snippets, tool calls and their results, as a team or public link.
They are available on Teams and Enterprise plans only, with a limit of 50 shares per day, and not with "No Storage" Privacy Mode. Cursor applies best-effort redaction of known secret patterns and warns that redaction can miss secrets, so review a transcript before you share it.
Which method should you use?#
Use WaveBrain as the running record of what Cursor's agent did, Git for the code itself, and checkpoints when you need to undo an edit in the current chat. Add hooks only if you need something specific that you are happy to maintain. For a broader look at the same question, read how to see what your AI agent did.
WaveBrain's free plan covers 1,000 actions a month with 7-day history and no credit card. Start free and paste the Cursor config from the Setup page. For every step, including how to check the first entry, follow how to connect Cursor to WaveBrain.
Frequently asked questions#
Does Cursor keep a history of agent changes?#
Cursor keeps checkpoints, which are local snapshots of files the agent modified during a chat. They let you preview and restore files, but Cursor says to use Git for permanent version control.
Does restoring a Cursor checkpoint delete the chat?#
No. Cursor's docs say restoring a checkpoint reverts files only and does not remove messages from the conversation.
Can I see what Cursor's agent did outside my code?#
Not from checkpoints or Git, which only cover files. Connecting Cursor to WaveBrain over MCP records deploys, pushes, messages and other outside actions on a timeline.
Where do I configure MCP servers in Cursor?#
Cursor reads .cursor/mcp.json in a project for project-specific servers and ~/.cursor/mcp.json in your home directory for servers available everywhere.
Are Cursor hooks available in cloud agents?#
Command-based hooks in your repository's .cursor/hooks.json run in cloud agents, but Cursor's docs say user-level hooks in ~/.cursor/hooks.json are not available there because cloud agent VMs cannot read your home directory.