Connect your knowledge base to your AI
Maverto speaks the Model Context Protocol. An AI assistant that speaks it too can read what your company knows — with your own permissions, and nothing beyond them.
Server address
https://api.maverto.com/mcp
This address is everything your assistant needs. There is no key to copy and no configuration file to fill in.
Connecting
Three steps, once
Every AI client words it differently, so this describes what happens rather than which button to press.
Step 1
Add the address
Wherever your AI lets you add an MCP server — the field that asks for a URL — paste the address above and save it.
Step 2
Sign in as yourself
Your browser opens a Maverto sign-in page. You sign in with your own account: the assistant never sees your password and never gets one of its own.
Step 3
Allow the access
Maverto shows what is being asked for and by whom. You pick the company whose knowledge base you are connecting and confirm. From then on the assistant can read — and only read.
The connection belongs to you, not to the assistant: it carries your account and your permissions, and it ends the moment you take it back.
What the consent screen shows
Before anything is granted, your browser shows a Maverto page that states in plain words what you are about to allow.
If the application or the return address is not the one you started yourself, deny it.
Which application is asking, and the address you will be returned to.
That it is asking to read your records — the only thing it can ask for.
The account you are signed in as, and the company you are connecting.
Two buttons: allow, or deny. Nothing is granted until you press one.
What it can do
Three tools, all of them read
This is the whole surface. There is nothing else behind it.
Search by meaning
Your assistant asks in plain language and gets the records that answer the question, not only the ones that repeat the same words. It can narrow the search to one workspace and one type of record.
Read a record in full
A SOP with its steps, a decision with its summary and date, a note, an update: the whole text with its status and its verification state, instead of a fragment out of context.
Read a video as text
Recorded walkthroughs come back as a transcript — full text, summary and timed segments — so an explanation somebody gave once on camera can answer a question today.
Boundaries
What it can never do
These are properties of the server, not settings you have to remember to switch on.
It only reads
There is no tool that creates, edits, publishes or deletes. An assistant connected this way cannot change a single record, and cannot be talked into it.
It has exactly your access
Every request is checked against your own permissions. A workspace you cannot open does not exist for the assistant either, and the workspace of a record is read from the record rather than from the request.
Captures from other screens never leave
Guides and shadow recordings — what the browser extension captured on third-party screens, with addresses, field names and typed-in values — are not published here. Not by search, not by type, not by a direct link to a known record.
Personal details are not published
Records come back without the name and the e-mail of their owner. A model that needs the text of a procedure has no reason to receive the people behind it.
Every read is on the record
Each call is written to the audit log as a view, marked as coming from the MCP server and naming the tool that caused it. The mark is set on our side, so a client can neither forge it nor erase it.
Taking the access back
The connection lives in your Maverto account, not inside the assistant. Open your account settings, find it in the list of issued access and revoke it. The next request the assistant makes comes back with nothing.
Revoking touches nothing else: not your data, not your team, not any other connection.
Your procedures, one address away
Put what your company knows into Maverto, and the AI you already use can answer from it — read-only, inside your permissions, and in the log.