We were nervous about pointing any AI at our real systems. The thing that changed our minds was being able to define exactly what the agent is allowed to reach. We scoped it to the specific tables that matter and left everything else out of view. That gave our security lead something concrete to sign off on rather than a vague promise, and it meant we could open the agent up to more of the team without worrying about what it might stumble into.

We added Peliqan as an MCP connection in Claude and it immediately had access to our business data. The agent can list our tables, look at the columns, and run the SQL it needs, so answers come straight from our systems instead of a guess. What we didn't expect was that it can also act on our apps, not just read them, so it updates a record where we need it to. Setup was an API key and the endpoint, and we were testing the same day.

We do data and AI work for several clients, so we needed a way to run separate setups without managing a different tool for each one. Peliqan let us keep client environments separate while using the same endpoint approach for each agent. For each client we connect their sources (one runs on Exact Online, another on Odoo), scope what the agent can read and write, and hand them an assistant that works in whatever AI client they already use. The fact that the same setup works in Claude and ChatGPT means we don't have to ask which tool the client prefers before we start.