Connect your AI

Let your Claude or ChatGPT-Codex use Loggia's tools — look up a customer, check the pipeline, tee up a change. Every write still asks you first. Reading happens straight away; anything that would change your data stops and waits in your approval queue until you tap approve.

Takes about five minutes: mint a key, paste one line into your AI, done.

Step 1

Make a key for your AI

The key is shown once, here, and never again — copy it before you leave this page. If you lose it, revoke it below and make another.

What is this key for?

Look, don't touch — this key can only read. It will not be able to propose changes at all, so nothing will reach your approval queue.

Step 2

Paste this into your AI

Make a key above and these fill in with it. Until then they show a placeholder you would have to replace by hand.

Run this in your terminal

Then start Claude Code and type /mcp to check it connected. To take it away again: claude mcp remove loggia.

What your AI will be able to do

Exactly the tools you have published in Agent Studio — nothing else in Loggia is reachable from a key. Publish more, and they appear to every connected AI on its next look.

Your connected AIs

Revoking is instant: the next thing that key tries stops at the door.

What stays true no matter which AI you connect

Every write asks you first. A tool that changes anything — adds a contact, tags one, moves a job, logs a note — never runs when your AI calls it. It becomes a request in your approval queue, and it only runs when a signed-in person in your workspace approves it. That gate is in Loggia's code, not in a setting on the key, so no key and no AI can turn it off.
Reads happen straight away. Lookups, filters and summaries return immediately — that is the point of connecting. A read-write key can also see contact emails and phone numbers in tools you published with contact details switched on.
It cannot message your customers. Keys made here are never granted the two separate permissions that let an agent propose a text, an email, or a drip enrolment. Even proposing one takes a key you mint deliberately in Agent Studio.
It only ever sees your workspace. A key is tied to this organisation. There is no request it can make that reaches another Loggia customer's data.
Everything is written down. Every call, refusal, queued request and approval lands in your Agent Studio receipt ledger, which is hash-chained — entries cannot be quietly edited after the fact.
There are limits. Calls are capped per tool per hour and across the whole workspace, and the approval queue itself is capped — so an AI that gets stuck in a loop stalls instead of flooding you.
Revoking is immediate and final. A revoked key cannot be un-revoked, and it stops working on the very next call.

Approval queue and receipts in Agent Studio