Preview before publishing
Every tool is runnable behind your login first. Publishing is a decision, not a leap.
Brand MCP Server · New
Your brand, callable by AI agents, not just readable
llms.txt is a document an agent has to find, fetch and interpret. An MCP server is an interface it can query. Same knowledge, published as tools an agent calls directly.
The shift
Crawling was for search engines. Agents ask questions.
Everything the web publishes for machines assumes a crawler: fetch the document, parse it, store it. Agents do not work that way. They have a question, a budget for context, and a preference for whatever answers it in one call.
A text file
The agent has to know it exists, fetch the whole thing, and hope the part that matters survives its context window. Everything you published competes with everything else you published.
A server
The agent asks its actual question and gets a scoped answer. What it needs, when it needs it, without spending its context on the rest of your site.
What gets published
Five read-only tools
Generated from the knowledge you already maintain in Targetlytics: the same material behind your llms.txt, your schemas and your public pages.
search_brand_knowledgeAnswer a question from your product and marketing knowledge.list_pagesEnumerate the pages you have made available.get_pageFetch one page as clean, structured content.get_llms_txtReturn your llms.txt without a separate fetch and parse.get_schemasReturn the structured data for a page, or for the site.
Read-only, and scoped to what you already publish. Nothing from your Targetlytics account is exposed. No analytics, no competitive data, no prompts. Turning it on is the publish step, the same trust model as putting llms.txt on your site.
How it works
One switch, then a URL
- 1
Try the tools
Run any tool from inside the app, behind your login, and see exactly what an agent would get back.
- 2
Publish
One switch makes the endpoint public. You get a connector URL and a snippet for your llms.txt.
- 3
Reference it
The llms.txt reference is the discovery signal that works today. Paste it and you are done.
Reversible
Switch it off and the endpoint stops responding. Nothing is baked into your site.
Stays in sync
It serves the same knowledge base your llms.txt and schemas are generated from, so it does not drift.
What it does and does not do yet
Agents do not auto-discover MCP servers yet
There is no discovery standard an agent crawls today, and anyone telling you otherwise is selling something.
The practical signal right now is referencing your endpoint from your llms.txt, which the product generates for you. A machine-readable descriptor served from your own domain is planned.
So why now? Because the cost is one switch, and the sites that became machine-readable early are the ones AI engines learned to trust. This is the same bet one protocol generation later, and it costs you an afternoon rather than a rebuild.
Not to be confused with
Publishing outward, versus querying inward
Brand MCP server
Publishes your brand knowledge outward, for other people's agents. This page.
Targetlytics MCP connector
Brings your Targetlytics data into your own assistant, for you. See the connector.
Frequently Asked Questions
Brand MCP server, explained
A Model Context Protocol server published for your brand. Instead of leaving an AI agent to find and parse text on your site, it exposes your product and marketing knowledge as a small set of tools the agent can call directly: search your knowledge, list your pages, fetch a page, read your llms.txt, read your schemas.
llms.txt is a text file an agent has to discover, fetch and interpret. An MCP server is an interface it can query. The same knowledge, but the agent asks a question and gets a scoped answer instead of downloading a document and hoping the relevant part survives its context window.
Not yet, and we will not pretend otherwise. There is no discovery standard an agent crawls today. The practical signal right now is referencing your MCP endpoint from your llms.txt, which the product generates for you, and a machine-readable .well-known descriptor on your own domain is planned.
It is read-only and scoped to the brand knowledge you already publish, meaning the same material behind your llms.txt, schemas and public pages. It exposes nothing from your Targetlytics account, no analytics, no competitive data. Turning it on is the publish step, the same trust model as putting llms.txt on your site.
Yes. Every tool can be run from inside the app, behind your login, so you can see exactly what an agent would get back before you make the endpoint public.
Because the cost is one switch and the position is hard to retrofit. Sites that were machine-readable early are the ones AI engines learned to trust. This is the same bet, one protocol generation later, and unlike most such bets it costs you an afternoon rather than a rebuild.
No, and they point in opposite directions. The connector brings your Targetlytics data into your assistant, for you. The brand MCP server publishes your brand knowledge outward, for other people’s agents.
Make your brand answerable, not just crawlable
Start with a free audit of how AI-readable your site is today.
Get a free visibility audit
