Beyond the Table API: Implementing Agentic AI for ServiceNow with Claude & MCP

Nick Zitzer1
Tera Contributor

3 hours ago

Most ServiceNow AI integrations today suffer from the same limitation: they treat the platform like a flat database. If you point a standard LLM at the Table API, it might be able to read an Incident, but it doesn't understand the architectural context of the CMDB, the nuances of the Task lifecycle, or the business impact of a Change Request.

To build true AI Agents for ServiceNow, we need to provide two distinct layers:

  1. The Action Layer (MCP): The secure bridge that allows the AI to execute operations.

  2. The Logic Layer (Skills): The "domain expertise" that teaches the AI how to be a ServiceNow Architect.

I’m excited to share two open-source projects aimed at standardizing this for the community.

1. The Action Layer: Node.js MCP Server

We’ve released a Model Context Protocol (MCP) server specifically for ServiceNow. Instead of writing custom integrations for every prompt, MCP allows Claude (and other MCP-compatible models) to discover ServiceNow tools dynamically.

  • Secure CRUD: Performs operations via standard ServiceNow APIs.

  • System Discovery: Allows the agent to understand the schema before attempting to write data.

  • Node.js based: Easily deployable and extensible for custom Scripted REST APIs.

GitHub: mcp-servicenow-nodejs

2. The Logic Layer: Claude Skills Library

Access without intelligence is dangerous. This library provides a set of "Skills"—structured logic and context—that teaches Claude how to navigate ServiceNow's complexities.

  • Incident Resolution Skills: Logic for deep-diving into logs and related tasks.

  • CMDB Auditing: Teaching the AI to look for "orphaned" CIs or broken relationships.

  • Agentic Thinking: Moving Claude from "answering questions" to "executing multi-step workflows."

GitHub: happy-servicenow-skills

How to use them together

You can use the Skills library Standalone within Claude Projects to act as a "Consultant" that advises you on code and architecture.

However, when you connect the Skills to the MCP Server, you get an Agent that can:

  1. Identify an issue in the CMDB (Skill).

  2. Query the affected records (MCP).

  3. Draft a remediation Change Request (Skill).

  4. Propose the update for your approval (MCP).

Why this matters for the Community

We are moving past "Chat with your data." We are entering the era of Agentic Operations. By open-sourcing these libraries, my goal is to help SN Developers stop rebuilding the "connectivity" wheel and start building the "intelligence" that actually reduces the backlog.

I’d love to get your feedback, PRs, and see how you’re implementing MCP in your instances!

  • 102 Views
0 REPLIES 0