Can ServiceNow expose its own skills, REST APIs, flows, and knowledge graphs as tools for an external AI to consume? The answer is yes. Watch as Sayan and Sujan walk though the setup and end to end testing. This is in depth technical! We start in the MCP Server console, create a brand-new MCP server from scratch, and attach tools to it for incident lookup and incident modification. From there we set up the inbound integration side: registering an application in the Machine Identity Console, generating a client ID and secret, and switching the token format to JWT so the connection is properly secured. Then we jump over to Claude, build a custom connector using that MCP URL and credentials, and connect it in a user-scoped way, meaning the external AI never gets more access than the logged-in user already has. Control in your hand. Once connected, we go further: exposing a custom scripted REST API as an MCP tool, refreshing the server's tool list, and watching the new tools show up automatically. Finally, we put it all to the test live by asking Claude in natural language to create an incident, add a work note to it and retrieve data through the custom REST API tool, all without touching the ServiceNow UI 0:00 - Recap: MCP as Client, and flipping the direction 0:56 - MCP Server console overview & requirements 1:32 - Creating a new MCP server 2:28 - Why you need multiple MCP servers 2:51 - Adding tools: incident lookup & modify incident 3:23 - Setting up the Machine Identity Console 4:12 - Token format, scoping & JWT setup 4:34 - Building the custom connector in Claude 5:27 - Connecting & user-scoped authentication 6:27 - Attaching a Now Assist skill to the MCP server 7:23 - Exposing a scripted REST API as a tool 8:28 - Registering the REST API tool 8:54 - Refreshing the tool list 9:19 - Live test: creating an incident via natural language 10:09 - Live test: adding a work note 10:53 - Live test: pulling data via the custom REST API tool