How it works
The Model Context Protocol is how modern AI clients call external tools. Building and operating that layer yourself means writing a server, handling auth, generating schemas, and answering for every action it takes. My MCPaaS does that part.
Your organization is the billing and security boundary. It comes with two isolated environments — development and production — so you can prove a change works before it touches anything real.
A connector describes one system: an IMAP mailbox, a REST API, a database. You provide the configuration and credentials once. Credentials are encrypted on save and never displayed again, to anyone, including us.
Before a connector is exposed to any AI, you run a test that proves the connection works and shows exactly which capabilities are available. After that, scheduled checks keep watching, and the last error is surfaced in the portal rather than buried in a log.
A server is a hosted endpoint that exposes a chosen set of connectors as tools. Tools are generated from connector capabilities and your policy rules — never hand-written per customer — so every tool has a stable name, typed input and output schemas, a read-or-write classification, and a timeout.
You see the full generated tool catalog before anything goes live. Naming collisions and policy conflicts are caught here rather than by a confused AI client at runtime.
Each AI client gets its own revocable, expiring credential. We store only a hash — the token itself is shown once, at issue. Usage is tracked per credential, so you can see which client is doing what.
Paste the endpoint into your AI client. It lists the tools, calls the ones you allowed, and every invocation lands in your audit trail with the actor, target system, outcome, and a correlation ID.