An MCP server, so an assistant can do the work
Lynkarr speaks the Model Context Protocol. Point Claude — or anything else that speaks MCP — at one URL, and the codes on your posters become something you can change by asking.
What MCP is, briefly
The Model Context Protocol is the standard way an AI assistant talks to a service that is not built into it. An MCP server publishes a set of tools; the assistant reads what each one does and calls the right one. It is the difference between an assistant that can describe your QR codes and one that can change them.
Ours is a remote MCP server, hosted by us at https://lynkarr.com/mcp. There is nothing to install, nothing to keep running, and no separate credential: it takes the same API token the REST API does.
The thing it is actually for
A poster is printed. The campaign it points at has moved. Somebody says so, and the code now goes somewhere else — no reprint, no hunting through a dashboard for the right record, no risk of editing the wrong one because two are named similarly.
That is the whole argument. Everything else an assistant can do here — listing codes, making them, reading how a campaign went — is convenience. Repointing something already in the world is the part that is genuinely hard to do any other way.
It cannot do more than a script could
Every tool is a call to the same public API, with the same token, the same scopes, the same per-minute rate limit and the same monthly quota. There is no privileged path in which the assistant is more capable than your own code would be, because there is no second implementation for one to appear in.
Scopes are the control, and they are per-token. An assistant given a read-only token is not shown the tools that would write — not shown, rather than refused, so it cannot try, fail, and report that something is broken.
What it will not do
It cannot tell you who scanned anything, because nothing here records that: no cookies, no stored addresses, no cross-site identifier. The answer is how many, from where and on what.
It cannot change your plan, your billing, your team or your account settings. It cannot read what people submit through your forms. And it cannot repoint a static code, because a static code's destination is in the printed pattern — the tool says so rather than appearing to succeed.
Getting there
-
Make an API token
Settings, then API tokens. Name it after the assistant, so you know what you are revoking later.
-
Choose its scopes
Read-only is codes:read, links:read and analytics:read. Add the write scopes only if you want the assistant to create and repoint.
-
Add the server
In your assistant, add a remote MCP server at https://lynkarr.com/mcp. Where a client offers only OAuth — Claude's connector dialog does — set its authentication to none and send the token as an X-API-Key header instead.
-
Ask it something
“List my QR codes” proves the connection. “Point the autumn poster at the new page” proves the point.
More in the same place
Questions
What is an MCP server?
MCP — the Model Context Protocol — is the standard way an AI assistant talks to a service it does not have built in. An MCP server publishes tools the assistant can call. Lynkarr's is a remote one at https://lynkarr.com/mcp: nothing to install, authenticated with a normal Lynkarr API token.
Which assistants can connect to it?
Anything that supports remote MCP servers over HTTP. Claude does, and the setting is usually called connectors, integrations or MCP servers. Because it is a standard rather than an integration we built for one product, a client that speaks MCP works without us doing anything.
The connector only offers OAuth. Where do I put the API token?
Set the connector's authentication to none, and send the token as a custom header called X-API-Key. Use that rather than Authorization: a client that manages OAuth itself tends to treat Authorization as its own and quietly drop a custom one, which leaves a connector that connects and then reports no tools. We accept either header. The step-by-step for Claude, ChatGPT, Copilot, Cursor and VS Code is in the support article.
Can an AI assistant change where my printed QR code points?
Yes, if you give its token the codes:write scope. That is the case worth connecting it for: the pattern on the paper does not change, so everything already printed keeps working and starts going somewhere new. Static codes cannot be repointed by anyone, including you, because the destination is in the pattern.
Is it safe to connect an assistant to my account?
It is exactly as capable as the token you give it, and no more — every tool is a call to the same public API with the same scopes and limits. Start with a read-only token. Three things no token can ever do, whatever scopes it holds: invite or remove a person, change a role, or touch your billing. Team access is read only by design, because those are what turn a leaked token into a lost account.
Can an assistant read what people submit through my forms?
Only with forms:read, and only if you grant it. That scope covers form submissions and the enquiries sent through a digital card, because reading the responses is most of what a form is for — so it is a meaningful thing to tick rather than a side effect of letting an assistant tidy up your form titles. Both are read only: nothing can delete somebody's enquiry.
Does the MCP server cost extra?
No. It is available on every plan including the free one, and it uses your normal API allowance rather than a separate one.
Try it with one code
The free plan has no expiry and asks for no card, and the analytics behind it are the same ones a paying customer gets.
Free forever for one person. No card, and scans are never metered.