See every new admin. Approve what goes live. Undo one change.
Nothing goes live on your site that you didn’t approve.
Don’t let AI touch your site without an undo button.
Lyntcast sits between your CMS and everything that writes to it: teammates, plugins, automations, AI agents, and attackers using a stolen login or a hacked plugin. Every change is diffed, checked against your rules and risk-scored. Safe ones go live, risky ones wait for a reviewer, and any of them can be undone.
No credit card. The demo runs in your browser with no account. Works with Claude, ChatGPT, Cursor, n8n or any MCP agent.
- WordPress
- WooCommerce
- Shopify(beta)
- Sanity
- Contentful
- Ghost
- Strapi
- Webflow(beta)
Agents, plugins and hackers are changing your site. Nobody is checking.
AI agents publish straight to production
An agent with a CMS token can rewrite a page, cut a price or delete a product in seconds. Most CMSs commit edits to published content immediately, so the first reviewer is your customer.
Hijacked agents and hacked accounts look the same as you
Text an agent reads online can tell it to add a link or a script. A stolen admin password writes spam into posts. To the CMS, both are just another edit.
When something goes wrong, nobody can say what happened
Who changed it, which tool, on whose instruction, and what did it say before? Restoring a full backup loses every good change made since.
Why people trust it
Your agents never see your CMS passwords
Agents connect to Lyntcast with their own key or OAuth token, which you can revoke instantly. Your CMS credentials are stored encrypted and added by Lyntcast’s server when it makes the call, so an agent can’t leak them or use them to go around your rules.
Alerts in minutes, sent from outside your site
When the WordPress plugin sees a new administrator or holds a change, Lyntcast posts to Slack straight away, and reviewers can get an email. Alerts come from Lyntcast’s servers, not your site’s mail, so they don’t depend on your site and can’t be deleted from it once sent.
No code on your public pages
The WordPress plugin adds no scripts or markup to the pages visitors see. Its content checks run when content is saved, not when pages load.
Undo any single change
A snapshot is taken before every write. Undo one change without restoring a backup; if someone edited the same fields since, Lyntcast shows you instead of overwriting their work.
Uninstalls cleanly
Deactivate the plugin and WordPress runs as before. Uninstalling it deletes everything it stored: settings, activity log, integrity baseline and per-post markers. Disconnect a site in Lyntcast and nothing more is written to it.
How it works
Step 1: Connect your CMS
WordPress connects in one click from its own “Authorize Application” screen. Other CMSs take one API token. Lyntcast checks it can write, and re-checks every hour.
Step 2: Choose your rules
Start with a ready-made pack (Essentials, E-commerce, Financial services, Health) and adjust: price limits, protected fields, banned claims, allowed link domains, plain-language AI policies.
Step 3: Point your AI agents at Lyntcast
Agents connect to one MCP link, by OAuth or with a key, instead of to your CMS. They never hold your CMS credentials, so they can’t go around the rules.
Then every change follows the same path: diff → rules → risk score → auto-approve or hold for review → apply with a snapshot → audit log. Undo is one click.
What the agent sees
An agent asks to cut a product price by 35% and adds “lowest price ever” to the description. Your Essentials rules say both need a person. The change is held, the CMS is untouched, and the agent is told exactly why, so it can tell the user or revise the change.
- • The reviewer gets a Slack alert with a direct link to the diff and preview.
- • Approve, and it’s applied to the product’s current state, with a snapshot for undo.
- • Reject with a note, and the agent reads the note from get_change_status.
{
"change_id": "chg_8k2m4q9xw1r7d3pz",
"status": "pending",
"outcome": "Pending review: a human must approve this change before it is applied. Do not retry it.",
"risk": "high",
"summary": "Edit regular_price, description on \"Linen shirt\"",
"document_id": "wc-products:412",
"findings": [
{
"rule": "Price changes over 20%",
"action": "require_approval",
"message": "\"regular_price\" fell -35.0% (40 → 26); limit is -20%"
},
{
"rule": "Risky marketing claims",
"action": "require_approval",
"message": "Adds banned phrase \"lowest price ever\" in \"description\""
}
],
"review_url": "https://lyntcast.example/w/acme/changes/chg_8k2m4q9xw1r7d3pz"
}Free plugin
A content guard for WordPress
Your security plugin or host firewall guards the door; keep it. Lyntcast guards the page. It adds no code to your public pages and only runs when content is saved. The plugin is configured automatically when you connect, starts in Monitor mode so you can watch for a week, then switches to Enforce.
Holds publishes
New posts and edits to live pages wait for approval, from any source, including administrator accounts.
Blocks injected code
Scripts, event handlers, javascript: links, hidden text, forms, unknown iframes and obfuscated code are refused as they’re written.
Quarantines rogue admins
Administrators created outside wp-admin are demoted on the spot and you’re alerted.
WooCommerce price guard
Price drops over your limit made outside Lyntcast are blocked; the EU 30-day “was” price rule uses recorded price history.
Security scan
Finds injected code already in your content, including spam shown only to search engines, and restores a clean version.
Small extras
Publishes “missed schedule” posts and can switch off XML-RPC.
Strapi, Contentful and Sanity
Content history, undo and an audit trail, without an Enterprise plan
Every change that goes through Lyntcast to Strapi, Contentful or Sanity is diffed field by field (including Strapi Blocks, Contentful Rich Text and Portable Text), saved with a snapshot, credited to the exact agent, person or app that made it, and can be undone on its own. That’s on every Lyntcast plan, including Free, whatever plan you have with the CMS. It isn’t a copy of their own versioning or release features: it covers the changes that go through Lyntcast, and CMS webhooks tell you about edits made elsewhere.
For agencies
Run every client site from one place, and let clients approve without an account
The Agency plan is $99 a month for 25 sites, then $3 per site. Reviewer seats are free, so your clients never need a paid seat. The first 50 agencies keep their launch price for as long as they stay subscribed.
Client approval links
Send a client a link to the changes you want signed off, with a preview and a deadline. They approve the exact version you sent, without an account or a login.
One view across every client site
See what’s waiting for review across all your client workspaces on one screen, instead of opening each one.
Approval routing
Send held changes to the right reviewer, such as the client who owns that site.
Monthly supervision reports
A PDF per client workspace: what agents proposed, what was held or blocked and what people decided, to send with your care plan.
Everything you need to let agents work, and to prove you’re in control
Field-by-field diffs
See exactly what changed, including inside rich text (HTML, Portable Text, Contentful Rich Text), plus a rendered before/after preview in a sandbox.
Risk scoring and auto-approval
Safe changes go live straight away. Risky ones wait in the review inbox. Rule-breaking ones are blocked, with reasons the agent can act on.
Rules agents read first
Agents fetch your rules in plain language and dry-run changes with check_change, so they fix their own work before anyone reviews it.
Undo one change or a whole afternoon
A snapshot is taken before every write. Undo a single change, or everything one agent did in a time window. Edits made since are never silently overwritten.
Conflict detection
Version checks refuse changes based on an outdated read. If two changes touch the same fields, the reviewer sees then / now / proposed and decides.
Scheduled changes
Go live later, and end automatically: a sale’s start and its end are approved together, and the old values come back on time.
Emergency stop and daily limits
One switch pauses every agent. Per-agent daily limits on changes and deletes, and automatic pause on a sudden burst or a run of rejections.
Site watch
Hourly checks of robots.txt, the sitemap and the home page as Google sees it: de-indexing, foreign sitemaps, injected spam pages and new outbound links.
Alerts where you work
Slack alerts name the change, its risk, who made it and why, with a direct review link. Reviewers can also get an email when changes are waiting.
Teams and roles
Owner, admin, reviewer and viewer roles. Reviewer seats are free on every plan. Approval routing and two-person approval are on the way for Business.
Tamper-evident audit log
Every proposal, decision, write and undo is recorded in a hash-chained log you can verify and export. Unlimited history on every plan.
Connection health
Every connection is tested for write access when you add it and hourly after. If a token stops working, you’re told what to fix in plain words.
Security details
Credentials encrypted at rest
CMS tokens are sealed with AES-256-GCM and only decrypted on the server to make a call. Agents never see them.
Standards-based agent sign-in
OAuth 2.1 with PKCE (S256 only), short-lived access tokens, rotating refresh tokens with replay detection, instant revocation.
Tamper-evident records
The audit log is hash-chained: altering or removing an entry breaks the chain, and an integrity check shows it.
No calls into private networks
Every outbound request to a CMS address, and every redirect hop, must resolve to a public address.
Questions
Does Lyntcast replace my CMS?
No. Your content stays in your CMS. Lyntcast reads it through the CMS’s own API, checks proposed changes, and writes approved ones back. Turn Lyntcast off and your site carries on as before.
Which AI agents work with it?
Any agent that speaks MCP over HTTP: Claude.ai, Claude Desktop, Claude Code, ChatGPT, Cursor, Windsurf, VS Code, n8n, and your own agents built with the Claude Agent SDK, OpenAI Agents SDK, LangChain or the Vercel AI SDK. Apps that support OAuth connect without a key.
Can an agent go around the rules?
Not through Lyntcast: agents never receive your CMS credentials, only a Lyntcast agent key or OAuth token that you can revoke instantly. If someone also gives an agent a direct CMS token, CMS webhooks (and the WordPress plugin) report edits made outside Lyntcast.
What about edits people make in WordPress directly?
With the free plugin in Enforce mode, new publishes and edits to live content are held for approval in Lyntcast, whoever makes them, including administrator accounts. Lyntcast’s own approved writes pass straight through.
Do I have to review every change?
No. Each agent has a trust level. Low-risk changes can go live automatically; only what your rules flag waits for a person. An adaptive level starts strict and earns more freedom after a run of clean approvals. How much needs a person depends on your rules and your agents, so we don’t quote a number: each workspace sees its own measured share of changes that needed a person, and which rules are worth relaxing.
Is my content sent to an AI model?
Only if you add a plain-language AI policy rule. Then the proposed change (not your whole site) is checked by a model through the Vercel AI Gateway. All other rules run on Lyntcast’s own servers without any AI.
Can it stop a hacker who controls my server?
No content tool can. Lyntcast protects what gets published, by whom, and how to undo it. Keep WordPress, plugins and themes updated and use strong passwords.
How much does it cost?
Free for one site, two agents and 500 changes a month. Paid plans start at $15 a month for one site; agencies get 25 sites for $99 a month. Reviewer seats and audit history are unlimited on every plan.
Let your agents work. Keep the final say.
Sign up and you get a demo store, the Essentials rules and an agent key with copy-paste setup. Connect a real site when you’re ready.