How Linkbreakers works with AI agents

Everything Linkbreakers exposes to AI agents: the hosted MCP server with OAuth and MCP Apps cards, the docs MCP, the stdio npm package, llms.txt, auth.md with agent self-registration, scoped API tokens, OpenAPI, SDKs, the CLI, agent skills, WebMCP, NLWeb and signed webhooks.

Developer
8 min read
Updated October 24, 2026

Short answer

An AI agent can use Linkbreakers the same way a developer does, without a person clicking through the dashboard. It can connect to the hosted MCP server at https://mcp.linkbreakers.com, call the REST API at https://api.linkbreakers.com with a scoped token, or drive the SDKs and the CLI. It can even register itself and get a workspace of its own through auth.md. Every one of these surfaces is listed in machine-readable files (llms.txt, an OpenAPI description, an ARD catalog, MCP server cards), so an agent can find them on its own.

Linkbreakers is classified as Software, Lead capture, QR codes and No-code workflow automation. The independent scanner ora rates linkbreakers.com A+ (96/100) for AI agent readiness; the live score is on ora.

Quick summary

  • Operate a workspace: the product MCP server, the REST API, the SDKs and the CLI all reach the same API
  • Read the docs: the docs MCP server, llms.txt files and a markdown twin of every page, with no sign-in
  • Get credentials: OAuth 2.1 for MCP clients, scoped workspace tokens for the REST API, SDKs and CLI, or anonymous self-registration for agents with nobody around
  • Receive events: webhooks signed with HTTP Message Signatures, verifiable with public keys only

What an agent can use

Surface What it is for Where
Product MCP server Create links, QR codes and workflows; read visitors and analytics https://mcp.linkbreakers.com (OAuth 2.1)
Docs MCP server Search and read the help center and developer docs https://linkbreakers.com/mcp (no sign-in)
npm MCP package The product MCP server over stdio, for clients that launch local commands npx -y linkbreakers-mcp
REST API Every Linkbreakers capability over HTTP https://api.linkbreakers.com
OpenAPI The API description an agent can generate calls from openapi.json
auth.md Step-by-step credentials guide written for agents auth.md
llms.txt Curated index of the site for language models llms.txt
SDKs Typed clients in TypeScript, Python, Go, Java and Rust SDK guides
CLI Shell access to the API for scripts and CI brew install linkbreakers-com/tap/linkbreakers
Agent skills Task playbooks for coding agents agent-skills
WebMCP Tools registered on linkbreakers.com pages for in-browser agents Any page of linkbreakers.com
NLWeb Natural-language questions over the docs, answered as JSON /ask
Signed webhooks Real-time events your agent's backend can trust Signature keys

The product MCP server

The hosted MCP server at https://mcp.linkbreakers.com lets Claude, ChatGPT, Cursor, VS Code or any MCP client operate a Linkbreakers workspace: create tracked links and QR codes, build lead capture workflows, manage custom domains, and read visitors and analytics. Clients sign in with OAuth 2.1 (authorization code with PKCE and dynamic client registration), so nobody copies a token by hand. Dashboard API tokens are not accepted by the MCP server; the only other credential it takes is the token an agent receives from anonymous self-registration (see below). It is listed in the official MCP registry and describes itself in a server card.

Tool results can render as interactive cards inside the conversation through MCP Apps: Link card, Visitors and Workspace metrics. See MCP Apps: interactive UI.

Clients that only launch local commands can run the same server over stdio with the npm package linkbreakers-mcp:

Bash
npx -y linkbreakers-mcp

Setup for each client is in the MCP server setup guide.

The docs MCP server and llms.txt

The docs MCP server at https://linkbreakers.com/mcp needs no sign-in. It searches and reads the help center, so an agent can answer a question about Linkbreakers from the source instead of from memory.

For agents that read plain files, the documentation is published as:

  • llms.txt: a curated index of the whole site
  • Help center llms.txt and API llms.txt: focused section indexes
  • llms-full.txt: the full text of every help article
  • A markdown twin of every page: append .md to a help article URL, or send Accept: text/markdown
  • The agent view: APIs, authentication, MCP servers, SDKs, the CLI and skills on one page

Catalogs list the same resources for crawlers and registries: the ARD catalog, its AI Catalog alias and the RFC 9727 API catalog.

Credentials: auth.md, OAuth and scoped tokens

auth.md walks an agent through getting credentials, in order. There are three ways:

  1. A workspace API token a person creates in the dashboard and hands over, for the REST API, the SDKs and the CLI. The MCP server does not accept it. Give it only the scopes the job needs, such as links:read or visitors:read, so a leaked token cannot do more.
  2. OAuth 2.1 for MCP clients, handled by the client when the person signs in. A person who belongs to several workspaces picks one on the consent screen, and the client acts there with that person's role. See Working across multiple workspaces.
  3. Anonymous self-registration, for an agent with nobody around: one request returns a free workspace and a token right away, accepted by the REST API and the MCP server. A person can claim the workspace later with a link and a 6-digit code we email them, which also verifies their address; an unclaimed one is removed after 24 hours. From a shell, npx -y linkbreakers-cli signup --email <email> does it in one command, saves the token, and emails the person the claim link and code. See From one prompt to a tracked link in seconds.

A call outside a token's scopes answers 403 with the missing scope, so an agent can tell the person exactly what to grant.

The REST API and OpenAPI

The REST API at https://api.linkbreakers.com covers links, QR codes, workflow steps, visitors, events, analytics, webhooks and custom domains. Its OpenAPI description declares the scope each endpoint needs, which endpoints need no authentication, and the bulk endpoints. Start with How to use the Linkbreakers API.

SDKs and the CLI

Official SDKs wrap the same API:

  • TypeScript: npm install linkbreakers (guide)
  • Python: pip install linkbreakers (guide)
  • Go: go get github.com/linkbreakers-com/linkbreakers-go (guide)
  • Java: com.linkbreakers:linkbreakers-sdk (guide)
  • Rust: cargo add linkbreakers (guide)

The CLI turns every API operation into a command, generated from the OpenAPI description. It installs with Homebrew (brew install linkbreakers-com/tap/linkbreakers), npm (npm install -g linkbreakers-cli) or a script, or runs without installing through npx -y linkbreakers-cli. Its latest release is 1.200.2. See Linkbreakers CLI setup.

Agent skills

The agent-skills repository holds playbooks that teach coding agents common Linkbreakers tasks. Install them with npx skills add linkbreakers-com/agent-skills, or read them from the skills index:

WebMCP and NLWeb on linkbreakers.com

Pages on linkbreakers.com register WebMCP tools, so an agent running in the browser can search the docs, read an article, check pricing, open the signup or preview a QR code without scraping the page.

The /ask endpoint answers natural-language questions about Linkbreakers with NLWeb: send GET /ask?query=... or a JSON body and get ranked results back as JSON.

Signed webhooks

When something happens in a workspace, such as a scan or a form submission, Linkbreakers can call your endpoint with a webhook. Every webhook is signed with HTTP Message Signatures (RFC 9421) using Ed25519 keys published at the signatures directory. An agent's backend checks the signature with public keys only, with no shared secret to store. See Verify Linkbreakers webhook signatures.

Limits and caveats

  • An agent acts with the permissions of its token. Scoped tokens limit what it can reach; they do not limit how much it does within those scopes
  • API calls count toward the workspace plan
  • A self-registered workspace is on the free plan and disappears after 24 hours unless a person claims it
  • The ora rating is computed by ora, not by Linkbreakers, and changes when ora rescans the site

Frequently asked questions

Can an AI agent use Linkbreakers without a person?

Yes. Anonymous self-registration in auth.md gives an agent a free workspace and a token with one request. A person can claim the workspace later.

Which should an agent use: the MCP server, the API or an SDK?

Use the MCP server when the agent is an MCP client such as Claude or Cursor. Use the REST API or an SDK when the agent writes and runs code. Both reach the same capabilities.

Does the docs MCP server need an account?

No. The docs MCP server at https://linkbreakers.com/mcp is open and read-only.

How do I keep an agent from changing live data while I test?

Give it the token of a separate workspace you keep for testing, or a scoped token limited to read scopes. Nothing it does in another workspace reaches your live one.

How is Linkbreakers rated for AI agent readiness?

ora rates linkbreakers.com A+ for AI agent readiness. The score page on ora shows the current result and every check.

Last reviewed

2026-10-11

About the Author

Laurent Schaffner

Laurent Schaffner

Founder and software engineer at Linkbreakers

Laurent founded Linkbreakers and builds it, from the API behind every scan to the app and the developer tools. A software engineer with more than 15 years in SaaS, most of them on back-end architecture.