Steps
To share an MLflow or TensorBoard dashboard with qURL™ by LayerV, publish TensorBoard on 127.0.0.1:6006 and send a qurl share link. For MLflow, allow the c-*.qurl.site Host, publish port 5000, then allow your teammate’s exact browser origin after they open the link. qURL keeps the tools on your machine and carries the live dashboard to your teammate.
Install the qURL CLI
Install the qURL CLI. On macOS, install it with Homebrew. On Linux and Windows, download the package for your platform from the latest release; the CLI page lists each one. Nothing asks you to sign up.
brew install layervai/tap/qurl qurl --versionTensorBoard: start it on localhost
TensorBoard serves on localhost port 6006 by default. Don’t pass --bind_all, which serves it on every interface. TensorBoard checks neither the Host nor the Origin header, so it needs nothing else for qURL, and the last step below is for MLflow only.
tensorboard --logdir runs --host 127.0.0.1 --port 6006MLflow: allow the host qURL sends, and keep localhost
MLflow 3.5.0 and later (mlflow server or its alias mlflow ui) rejects a Host it doesn’t know with 403 “Invalid Host header - possible DNS rebinding attack detected”. Through qURL the Host is c-….qurl.site. --allowed-hosts replaces the default list rather than adding to it, so list localhost and 127.0.0.1 again. MLflow uses fnmatch patterns against the complete Host header, including any port. The list below covers bare hosts and hosts with a port. A pattern is fine here: it names the host MLflow receives, not an origin it trusts.
mlflow server --host 127.0.0.1 --port 5000 \ --allowed-hosts "localhost,localhost:*,127.0.0.1,127.0.0.1:*,c-*.qurl.site,c-*.qurl.site:*"Publish the dashboard
Keep MLflow or TensorBoard running, and in a second terminal publish only the one you started: publishing a port with nothing listening fails. The first publish creates this device’s identity, starts a background service that keeps serving after you close the terminal (while you stay logged in), and prints the app’s CRID, its stable ID. The Access row says who can open the app: a plain publish makes it private from qurl 3.4.1 on.
qurl shareturns the CRID into an access link, and the link is what you send. Each dashboard is its own resource with its own CRID.qurl publish http://127.0.0.1:6006 # TensorBoard qurl publish http://127.0.0.1:5000 # MLflowCreate a link
qurl shareprints an access link (https://qurl.link/#…). On a terminal, the lines below it say who published the resource and when the link expires, for example “Expires in 1d (single use)”. A publisher LayerV has verified is shown as verified by LayerV; any other name is marked UNVERIFIED. With no flags the link can be opened once and expires within 24 hours. If the expiry line does not say “single use”, don’t send the link. Add --ttl 1h for a shorter window.qurl share <CRID>Send the link to one person
Send the qURL link directly to the recipient. Send it in a direct message, not a public channel: whoever opens it first uses it. Avoid email and apps that preview links, because a scanner that runs the page could use its one opening. Don’t open it yourself to check it, for the same reason. Run
qurl shareagain whenever you need a fresh link.MLflow: trust your teammate’s exact origin
MLflow blocks POST requests to its API from origins it doesn’t list, and its UI loads the runs table with a POST, so through qURL the experiment list appears but its runs fail with “Cross-origin request blocked”. Ask your teammate to send you the exact https origin from the address bar of the page they just opened (scheme and hostname, without a path). They must keep that page open; do not open their single-use link yourself. Restart mlflow server with that origin in --cors-allowed-origins and ask them to reload the same page. Do not use https://*.qurl.site: it trusts apps run by other publishers. Keep the security middleware enabled; do not use --disable-security-middleware.
mlflow server --host 127.0.0.1 --port 5000 \ --allowed-hosts "localhost,localhost:*,127.0.0.1,127.0.0.1:*,c-*.qurl.site,c-*.qurl.site:*" \ --cors-allowed-origins "https://r_your-link.qurl.site"End the review and restore local settings
Stop sharing, remove the recipient origins and qURL Host entries you added to the tool’s settings, then restart it with its security checks still enabled.
qurl stop <CRID>
What the recipient sees
Your teammate lands on the dashboard at an address like https://r_….qurl.site/.
More
In TensorBoard they see your scalars, graphs and images as soon as the link opens. In MLflow, after the reload, they browse experiments, compare runs and open artifacts. Neither tool has a sign-in by default, so they can also do whatever the UI allows, such as editing run names and tags in MLflow.
Troubleshooting
| If you see | Do this |
|---|---|
| MLflow answers 403 “Invalid Host header - possible DNS rebinding attack detected” | Use the --allowed-hosts list above, including c-*.qurl.site and c-*.qurl.site:* for a Host with a port. The list replaces MLflow’s defaults, so keep both bare and port-qualified entries for localhost and 127.0.0.1. |
| MLflow shows experiments, but the runs table fails with “Cross-origin request blocked” | Your teammate’s exact origin is missing from --cors-allowed-origins. Add the https://r_….qurl.site origin from their address bar, restart mlflow server, and have them reload. |
Show 7 more fixes
| If you see | Do this |
|---|---|
| mlflow server exits with a UsageError about security middleware parameters | The security flags work only with the default uvicorn server. Remove --gunicorn-opts or --waitress-opts. |
| qurl publish says the local app is not accepting TCP connections | Start the app first and check that curl works on the same address. qurl publish connects to localhost as 127.0.0.1, so if your app listens only on IPv6 (on macOS or Linux, lsof -iTCP -sTCP:LISTEN -nP shows [::1]), publish 'http://[::1]:PORT' instead, or start the dev server with --host 127.0.0.1 and publish that address. Neither requires listening on your LAN. Keep the quotes: zsh treats the brackets as a pattern. |
| The recipient sees “Access link invalid” | The link was already used or has expired. Run qurl share again and send the new link. |
| The recipient sees “Resource not found” | The resource was deleted after the link was made. Publish the app again, then run qurl share and send the new link. |
| publish rejects http://127.0.0.1 with “only HTTPS URLs are allowed”, asks you to log in, or prints no Access: line | Your qURL CLI is out of date. On macOS run brew update, then brew upgrade qurl; elsewhere install the latest release. After updating, qurl --version prints 3.4.1 or later, and every publish is private by default. |
| publish says this Linux session “can’t reach a systemd user manager” | You are in a container or sandbox without a user service manager. Run qurl publish with --foreground; sharing lasts as long as that process runs. Container and CI setup → |
| publish says “Quota Exceeded” with “protected resource limit reached”, or “Your qURL account has reached a plan limit” | You can see this without an account: a device with no account can keep up to 3 active resources, with links that last up to 24 hours. Run qurl list --status active, then qurl delete <CRID> for any you no longer need (it asks you to confirm; --yes skips that). qurl stop keeps the resource, so it does not free a slot. |
Frequently Asked Questions
What happened to TensorBoard.dev?
TensorBoard.dev shut down on January 1, 2024, and the tensorboard dev upload command no longer uploads anything. TensorBoard itself still runs locally, which is what a qURL link shares.
Does TensorBoard have a login?
No. TensorBoard’s maintainers have said they have no near-term plans to add authentication and suggest an authenticating reverse proxy. Keep it on localhost and share it by link, rather than with --bind_all.
Which MLflow version should I share?
Use an up-to-date patched MLflow release; this guide was reviewed against 3.17.2. CISA’s Known Exploited Vulnerabilities catalog lists CVE-2026-64849, a server-side request forgery in an unauthenticated MLflow webhook test endpoint. Its original fix shipped in 3.15.0, with follow-up IPv6 hardening in 3.16.0. The Host and origin checks arrived in 3.5.0; that is a feature minimum, not a safe version recommendation.
Who else can reach my app while the link is open?
Opening a qURL link starts a session for that browser, which lasts 1 hour by default. After a browser opens it, another browser on the same public IP cannot use its access. A script or agent that opens the link without a browser is admitted by its network address instead, which other devices on that public IP share until the session ends. A link admits whoever opens it: the API’s IP and country rules do not apply to the signed links qurl share issues. Keep your app’s own sign-in for anything sensitive.
Does my computer have to stay on?
Yes. Your MLflow or TensorBoard app runs on your machine. The machine and app must stay running and connected; the app is unavailable while the machine is asleep.
How do I stop sharing?
Run qurl stop <CRID> to pause serving the app, and qurl start <CRID> to resume it; a stopped app still counts toward your resource limit. qurl delete <CRID> removes the resource for good after asking you to confirm; its CRID can never be shared again.