Back to Projects
InternalInternal — Enterprise

ServiceNow AI Assistant

Agentic IT Support with Bedrock AgentCore

An AI assistant that manages ServiceNow incidents through conversational AI, powered by agentic tool-use orchestration with persistent memory. Deployed as both an embedded portal widget and standalone chat interface. Built 56 MCP tools across 7 IT service domains, going from 0 to 26 wired tools in a single day.

IT staff were navigating complex ServiceNow interfaces to manage incidents. The organisation needed a conversational layer that could list, search, and create incidents through natural language without replacing the underlying ticketing system.

PythonAWS BedrockAgentCoreChainlitFastAPIServiceNow

Designed the hardest piece: the dual-token auth architecture bridging two identity providers. Built the agent pipeline, MCP server with 56 tools, and the embedded portal widget.

56
MCP Tools
7
Domains Covered
2
Deploy Modes

Highlights

01

56 MCP tools across 7 domains

Built an MCP server from 0 to 26 wired tools in one day, expanded to 56 across incidents, changes, knowledge, users, groups, SLAs, and CMDB. Full agentic tool-use pipeline, not simple prompt-and-respond

02

Persistent conversation memory

Per-user session memory with 7-day expiry. System prompt dynamically includes the last 5 conversation turns for context continuity across messages

03

Dual-token authentication

PKCE flow issues tokens validated via JWKS. A token exchange layer converts identity tokens (RS256) into application JWTs (HS256) for the chat UI, with audience claim validation at every layer

04

Dual deployment interfaces

Embedded as a widget in the employee portal and as a standalone chat application. Both share the same FastAPI backend with SSE streaming and full markdown rendering

05

Tech Stack

AI & Agent:AWS BedrockAgentCoreStrands Agent FrameworkClaude Sonnet
Backend:FastAPIChainlitPythonhttpx
Auth:OIDCJWT (RS256/HS256)JWKS
Integration:ServiceNow APISocket.IOSSE Streaming