Hermes Agent Cursor import scores 8.3 out of 10 from me: it moves rules, MCP servers and skills cleanly, and only loses points on rule scoping.
The feature is hermes import-agent cursor, part of the agent-importers plugin that Nous Research maintains.
Every rating below is my editorial call from the official README and docs, criterion by criterion.
If you want help working out where agents fit your marketing, book a free SEO strategy session.
Hermes Agent Cursor import scorecard
| Criterion | Score | Why |
|---|---|---|
| Setup | 9/10 | hermes setup spots ~/.cursor and offers the import on a fresh install. |
| Rules carry-over | 7/10 | Rule bodies arrive; .mdc frontmatter, globs included, is stripped. |
| MCP servers | 8/10 | Servers move into config.yaml; secret-looking values are stripped and listed. |
| Skills | 8/10 | Category folders are allowed, but same-name skills clash. |
| Safety | 9/10 | Preview always, dry run on request, credentials never read, memory backed up. |
| Side-by-side sync | 9/10 | A digest-based --sync that the docs call safe to schedule. |
| Docs clarity | 8/10 | Clear tables, with one small mismatch on rule paths. |
| Overall | 8.3/10 | Average of the seven scores. |
How I scored it: each criterion is judged on what the README and the import guide say the importer does, not on speed or any test run.
A 10 means nothing is lost and nothing needs fixing by hand.
A 7 means it works, but a typical user has real tidying to do.
What the Cursor importer is
It is one command inside an official plugin, not a separate app.
The plugin used to be part of Hermes core and now ships from the hermes-official-plugins repo under an MIT licence.
Cursor support arrived with Hermes Agent v0.22.4 on 21 October 2026, where the notes describe agent-importers as "now with Cursor".
It reads ~/.cursor by default and writes into your active Hermes profile.
Under the hood, the command and its flags are the same ones Hermes used when the importer lived in core.
Upgraders keep their existing import-sync.json manifest, so nothing has to be set up again.
It is also narrow by design.
The Cursor mapping covers four inputs: AGENTS.md, rule files, mcp.json and skill folders.
Anything else in ~/.cursor is outside its scope.
Setup, rated 9/10
Getting started is close to effortless on a fresh install.
Setup checks for agent folders before anything else, names what it found, and installs the plugin if you say yes.
On an existing install it is one line, hermes plugins install agent-importers.
The missing point is the version trap: on a Hermes build with the old importer built in, the plugin stays inert and Cursor is not offered.
Brand-new installs get the smoothest version.
When several agent folders are found, setup shows a checklist with all of them ticked, so you can pick only Cursor.
Already set up? The manual route takes about as many commands as there are steps on a recipe card: install, dry run, import.
Version check aside, I like that setup never installs anything you did not agree to.
Say no, or have no agent folders, and nothing is installed at all.
Rules carry-over, rated 7/10
This is the weakest score, and it is a design choice rather than a bug.
AGENTS.md and every User Rule file become memory entries in MEMORY.md.
Cursor's .mdc frontmatter (description, globs, alwaysApply) is stripped, so only the body survives.
Hermes memory has no file-pattern switch, so a rule scoped to one folder in Cursor becomes general guidance.
Users with heavily scoped rules will have tidying to do.
Memory merging works well.
Entries are deduplicated against what is already in MEMORY.md, so a second import does not double your rules.
If you mostly use plain .md rules without globs, this criterion is effectively a 9 for you.
The fix is quick: add a line to each scoped rule in MEMORY.md saying where it applies.
🔥 Want the Hermes setup I would rate a 10 out of the box? Inside the AI Profit Boardroom you get step-by-step video trainings, the installable Agent OS, four coaching calls a week and 3,400+ members comparing agent setups. → Join the AI Profit Boardroom
MCP servers, rated 8/10
Every server in mcp.json under mcpServers lands in mcp_servers in config.yaml.
Environment variables and headers that look like secrets, such as *_TOKEN, *_API_KEY and Authorization, are stripped.
The report lists each stripped value, which is the right call.
It still costs a point, because every keyed server is broken until you re-add its secret.
A malformed mcp.json does not sink the run either; it becomes a per-item error while everything else imports.
That resilience is why this is an 8 and not a 7.
Servers that already exist in Hermes are reported as conflicts and skipped, so your working Hermes tools are never replaced by surprise.
Skills, rated 8/10
Skills import from skills/**/<name>/, so nested category folders are fine.
Each skill lands under ~/.hermes/skills/cursor-imports/, keyed by the folder that holds SKILL.md.
Flattening the categories means two same-name skills become a conflict, and existing Hermes skills win unless you pass --overwrite.
That is predictable, which matters more to me than clever merging.
Sync handles skills sensibly too: previously imported skills refresh in place when the Cursor copy changes.
If you edited the imported copy in Hermes, it is treated as a conflict and left alone.
Expect to rename a skill or two if you used category folders heavily in Cursor.
Safety, rated 9/10
- Every run prints a per-item plan before it writes.
--dry-runnever touches disk.~/.cursor/cli-config.jsonis never read.- The old
MEMORY.mdis backed up with a timestamp before memory is merged. - An unreadable
config.yamlis never rewritten, and your comments survive writes. - Non-interactive runs stop at the preview unless you add
--yes.
I would only call it a 10 with a one-command undo for the config side.
The safety score is the main reason I would let a junior team member run this without supervision.
The worst realistic outcome of a careless run is a conflict list, not lost work.
Side-by-side sync, rated 9/10
This is the best part for anyone who keeps Cursor open.
Each import is registered in import-sync.json, and hermes import-agent --sync re-imports only sources whose files changed.
Skills you edited inside Hermes are left alone, and a failed run keeps the old digest so the next sync retries.
The docs suggest pairing it with a daily Hermes cron job, which is how I would run it.
Profiles are well covered: a profile cloned with --sync-imports keeps pulling from the same Cursor tree.
Credential files never trigger a sync either, so token refreshes do not cause pointless runs.
My advice is to score your own setup against this criterion before you start: count the .mdc files that use globs.
Strengths and limits at a glance
| Strengths | Limits |
|---|---|
| Preview before every write | Glob and alwaysApply scoping is dropped |
| Credentials never read | Keyed MCP servers need secrets re-added |
| Sync that only touches changed files | Category folders are flattened, so names can clash |
Per-profile imports with -p | Older Hermes builds ignore the plugin |
| Malformed files fail one item, not the run | No permission or approval rules come across |
Read the left column as reasons to trust it, and the right column as your post-import checklist.
The missing point: sync works per profile, so each client profile needs its own manifest or a clone made with --sync-imports.
Docs clarity, rated 8/10
The import guide and the plugin README agree on almost everything.
The one gap: the guide lists rule files as rules/*.md, while the README lists rules/**/*.md, which includes nested folders.
I would trust the README, since it ships with the code, and check the dry run to be sure.
The guide also documents the upgrade path well: profiles with a registered sync source get the plugin automatically on update.
Small gap, solid docs, clear 8.
Each limit has a fix in under ten minutes, which is why none of them drags the overall score below 8.
Each of those three cases is about what is in your Cursor folder today, not about how big your team is.
📞 Want an expert verdict on your own SEO setup? Book a free SEO strategy session and my team will tell you which jobs an agent can take off your plate. → Claim your free session
Who should run it, and who should not
| Run it if you… | Skip it if you… |
|---|---|
| Keep rules and MCP servers in Cursor you want Hermes to follow | Have an empty or near-empty ~/.cursor |
| Plan to use Cursor and Hermes side by side | Depend on glob-scoped rules you cannot rewrite |
| Want scheduled jobs Hermes can run on a timer | Are stuck on a Hermes build older than v0.22.4 |
If you sit in the middle, run the dry run anyway.
It costs nothing, writes nothing and shows you exactly where your setup lands on this scorecard.
Teams in the middle, with a few scoped rules and one or two keyed servers, should still run it.
The tidy-up is small compared with rebuilding a setup by hand.
If you sit in the left column on two or more rows, the import pays off quickly.
Versus the Claude Code import
My Hermes import agent review covers the Claude Code and Codex paths.
Claude Code scores higher on permissions, because its allow and deny rules map into Hermes approvals.
Cursor scores higher on organisation, because it brings a whole rules folder and category-grouped skills.
Both share the same preview-first safety model.
Claude Code also skips its slash commands, while Cursor has no equivalent gap in its mapping.
Against doing it by hand, there is no contest.
Copying rules, MCP entries and skill folders manually gives you no preview, no dedup, no stripped-secret list and no sync.
Run it on a spare profile first if you are cautious, using -p, and compare the result with your main profile side by side.
Final verdict
8.3 out of 10, and a clear yes for any Cursor user trying Hermes.
The import is safe, inspectable and repeatable, and the sync turns it into a bridge rather than a one-off move.
Budget ten minutes to tidy scoped rules and re-add MCP keys, then run it.
For the full Agent OS and walkthroughs, join the AI Profit Boardroom.
Also On Our Network
- 🌐 The six-step run order for importing a Cursor setup into Hermes
- 🌐 Whether a small business should move its Cursor setup into Hermes
- 🌐 Five copy-paste prompts that get your agent importing Cursor
- 🌐 Which AI models to run after a Cursor import, with live GoldieBench scores
FAQ: Hermes Agent Cursor Import Verdict
What is the weakest part of the Cursor import?
Rule scoping. The .mdc frontmatter, globs included, is stripped, so scoped Cursor rules become general memory entries in Hermes.
Is the Cursor import safe to try?
Yes. It always previews first, --dry-run never writes, Cursor's credential file is never read and MEMORY.md is backed up before merging.
Does the Cursor import bring permission rules?
No. The Cursor mapping covers AGENTS.md, rules, MCP servers and skills only, so you set Hermes approvals yourself.
Which Hermes version supports Cursor imports?
Hermes Agent v0.22.4, released 21 October 2026, lists agent-importers as now supporting Cursor.
How is the overall score worked out?
It is the plain average of seven editorial scores, from setup to docs clarity, which comes to 8.3 out of 10.
📺 Video notes + links to the tools 👉
🎥 Learn how I make these videos 👉
🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉
About Julian
I'm Julian Goldie, an SEO entrepreneur, author and founder of the AI Profit Boardroom, which has 3,400+ members.
I help business owners scale with AI agents, automation and SEO.
- I've built a 7-figure SEO and link building agency, Goldie Agency, with a team of 70+ SEO experts.
- I've grown a YouTube channel to 400,000+ subscribers.
- I wrote "Link Building Mastery".
- My Udemy courses have taught over 29,000 students.
→ Get my best AI training inside the AI Profit Boardroom
Run the dry run, tidy your scoped rules, and if you want help planning the move, book a free SEO strategy session or join the AI Profit Boardroom for the full Hermes Agent Cursor walkthrough.
