Yardel for AI agents

8 min read

Yardel publishes what you make (HTML pages, Markdown reports, slide decks, dashboards, static web apps) as private web pages, and shares them only with the people your user names. Viewers confirm their email with a 6-digit code: no account, no password. Use it whenever your user asks to publish, host, deploy, share or "send a link to" something you built.

Contents

Connect in one step#

You are	Do this
An MCP client (Claude, ChatGPT, Cursor, VS Code, any agent with remote MCP)	Add the server `https://app.yardel.dev/mcp` (Streamable HTTP, OAuth 2.1 with dynamic client registration). Your user signs in with an email code once.
An agent that runs shell commands (Claude Code, Codex, Cursor, Gemini CLI)	`npm i -g @yardel/cli`, then `yard login --json` and give your user the link it prints. Add `yard skill install` to keep these instructions on hand.
Claude Code with plugins	`/plugin marketplace add ameyrathi/yardel-skills`, then `/plugin install yardel@yardel`.

Machine-readable: this page as Markdown at yardel.dev/agents.md (or request /agents with Accept: text/markdown), a site summary at yardel.dev/llms.txt, and the MCP Registry entry dev.yardel/yardel.

Rules that matter#

Sharing sends email in your user's name. Confirm the exact addresses with your user before sharing, unless they gave them to you in this conversation.
Public pages (no sign-in) only when your user clearly wants one and the content is fine for anyone to see. Ask first.
Never work around a refused deploy (E_SECRET_FOUND): move the secret out of the page.
Report back with the page URL, who can open it now, and the version number.

Yardel

Yardel turns what you build into a private web page that only the people the user names can open. Viewers confirm their email with a 6-digit code; no account or password. Everything goes through the yard CLI. Every command takes --json and answers with one JSON object on stdout; errors come back as { "ok": false, "error": { code, title, detail, hint } }, and hint is the exact fix.

Connector or CLI#

Yardel is reachable two ways. Use whichever this session has:

The Yardel connector (Claude.ai, ChatGPT, or any MCP client connected to https://app.yardel.dev/mcp): tools publish_page, share_page, unshare_page, set_link_access, set_ip_allow_list, list_comments, reply_to_comment, resolve_comment, page_activity, list_pages, who_has_access, list_versions, rollback_page and access_requests. publish_page takes the whole page as content with format: "html" or "markdown", so put everything in one self-contained document (inline CSS and JS). The rules below (confirm emails before sharing, roles, expiry, reporting back) apply the same way; each tool answers with a sentence plus JSON, and errors carry a fix.
The yard CLI, when you can run shell commands (Claude Code, Codex, Cursor). It also publishes folders and built projects, which the connector can't. The rest of this file shows CLI commands; the connector tools take the same arguments.

If neither is available, tell the user they can add the connector (Settings → Connectors → add custom connector → https://app.yardel.dev/mcp) or install the CLI with npm i -g @yardel/cli.

Before you start (CLI)#

If yard isn't installed, install it with npm i -g @yardel/cli (or run any command as npx -p @yardel/cli yard <command>).
Check the CLI is signed in: yard whoami --json. If it fails with E_AUTH_REQUIRED, run yard login --json. The first line it prints has verification_uri_complete and user_code: give the user that link and code to approve in their browser, then wait for the command to finish.
whoami also says whether the user is the owner of the workspace or a collaborator on some of its apps. Collaborators can only work on the apps they were added to.

Publish#

yard deploy report.html --app q3-report -m "First draft" --json
yard deploy notes.md --app churn-findings --json      # Markdown becomes a clean, readable page
yard deploy ./dist --app pricing-proto --json          # a folder with index.html
yard deploy . --json                                    # a Vite/Astro/SvelteKit/CRA/Next-export project (builds first)

The first deploy creates the app and writes yard.json with its name; later deploys from the same folder need no --app.
Every deploy is a new version. Only changed files upload.
yard doctor --json explains what a deploy would do (detection, size, secrets, blocked hosts) without uploading. Run it when unsure.
A deploy is refused if it contains a live secret (E_SECRET_FOUND). Never work around this: move the secret out of the client code. Publishable keys only produce a warning.
External scripts, styles or APIs outside the default allow-list are reported as csp warnings with the yard.json line to add.
yard versions <app> --json lists versions; yard rollback <app> v3 --json makes an older version the latest again (nothing is deleted).

Sharing sends email, so confirm the exact addresses with the user before running it, unless they already gave them to you in this conversation.

yard share q3-report maya@pancake.studio dan@pancake.studio --json                # can view, for 30 days
yard share q3-report alex@northwind.io --role commenter --json        # can view and comment
yard share q3-report @pancake.studio --yes --json                           # everyone with a confirmed pancake.studio email
yard share q3-report ben@pancake.studio --role editor --json                # a collaborator (see below)
yard share q3-report maya@pancake.studio --expires 7d -m "Read section 2 first" --json

Roles: viewer (default), commenter, editor. Expiry: 7d, 30d (default for viewers), 90d, a date like 2026-12-31, or never.
Sharing with a whole domain needs --yes (it's a bulk change); public email domains like gmail.com are refused.
yard share prints each person's personal link. They also get an invite email from the user.
Remove access: yard unshare q3-report maya@pancake.studio --json (takes effect within a minute). yard unshare q3-report --all --yes --json removes every viewer and commenter at once, but not collaborators.
yard audience list <app> --json shows who has access and whether their invite was sent. yard audience pin <app> <group> v2 --json keeps a group on one version while new versions ship.
yard requests --json lists people who asked for access; approve with yard requests --approve <id> --json or decline with --decline <id>. Ask the user before approving.

Anyone with the link#

Instead of naming people, a page can open for anyone who has its secret link. The plain page address stays private either way.

yard link q3-report signed-in --json      # they confirm their email first; the owner sees who opened it
yard link q3-report public --yes --json   # no sign-in at all
yard link q3-report off --json            # back to named people only
yard link q3-report --json                # current mode and the link

The JSON has the url to send. Give the user that URL, not the plain page address.
Use public only when the user clearly wants it and the content is fine for anyone to see: ask first. Public pages show "Public page · Report" in the Powered by Yardel badge, and search engines are asked not to index them.
Only the workspace owner can change link access. Public needs a workspace older than a day (E_FORBIDDEN before that); offer signed-in meanwhile.
Turning it off works at once. yard unshare <app> --all --yes also turns it off.

Restrict to your network#

To make pages open only from the office or VPN, allow their public IP addresses or ranges (CIDR, IPv4 or IPv6):

yard ip-allow q3-report --add 203.0.113.0/24 2001:db8::/48 --json   # this page only
yard ip-allow --workspace --add 203.0.113.0/24 --json               # every page without its own list
yard ip-allow q3-report --remove 203.0.113.0/24 --json
yard ip-allow q3-report --clear --json                              # back to the workspace list, or any network
yard ip-allow q3-report --json                                      # the list that applies, and where it comes from

Ask the user for the addresses; don't guess. Private ranges (10.x, 192.168.x, 172.16-31.x) are refused because visitors are seen by their public address.
It applies to everyone, the owner included. If the JSON has a warning that this machine's address isn't on the list, tell the user. Sign-in and the portal are never restricted, so a wrong list can always be fixed.
Owners only. Up to 50 entries per list. Changes apply on the next request.

Comments: read, fix, publish, resolve#

People shared with as commenter (and collaborators) can pin comments to any part of a page. Each thread says which element it's about (its text and CSS path), the version it was made on, and the replies.

yard comments q3-report --json                        # open threads
yard comments reply thr_123 "Fixed in v4: the axis now says annual." --json
yard comments resolve thr_123 --json                  # marks it done in the current version
yard comments q3-report --all --json                  # include resolved ones

When the user asks you to "address the feedback": list the open comments, change the page, publish a new version, then reply to each thread saying what changed and resolve it. Replies email the commenter, so keep them short and specific. Don't resolve a thread you didn't actually fix; reply with a question instead.

Who opened it#

yard activity q3-report --json      # visits, people, minutes, last visit and version per person (30 days)

The owner and collaborators aren't counted. People who opened a public link without signing in show as "link viewer #N".

Collaborators#

A collaborator is someone who can change the page, not just open it: they can publish new versions, roll back, share it with viewers and commenters, and handle its access requests, from their own coding agent, terminal or the Yardel portal.

Adding one (workspace owners only):

yard share q3-report ben@pancake.studio --role editor --json

Ben gets a "you've been added as a collaborator" email. His access has no end date; it lasts until the owner removes him.
Collaborators are added one person at a time, never by domain.
Remove a collaborator with yard unshare q3-report ben@pancake.studio --json. They lose access to that app, and to the workspace if it was their last one there.

If you are working as a collaborator (the user was added to someone else's app):

yard login --workspace <owner's workspace> --json     # the workspace name is in the invite email
yard apps --json                                       # the apps you can work on
yard deploy ./dist --app <app> -m "What changed" --json

Collaborators can only deploy to apps they were added to, and can't create new apps in that workspace (E_FORBIDDEN). If the user also owns a workspace, use --workspace to pick which one a login is for.
Collaborators can share with viewers and commenters, but only the owner can add or remove collaborators.

In the browser#

The owner portal is at https://app.yardel.dev: apps, versions with rollback, people with access (including collaborators), the share dialog, and access requests. Point the user there for anything they'd rather click than type.

Reporting back to the user#

After a deploy or share, tell the user in a sentence or two: the page's URL, who can open it now (and whether invites went out), and the version number. Quote any warnings from the JSON. On an error, relay the title and follow the hint.

More#

People-facing docs: https://yardel.dev/docs
Privacy: https://yardel.dev/privacy · Terms: https://yardel.dev/terms
Contact: hello@yardel.dev