ServiceNow MCP Server Setup: Step-by-Step Guide for 2026

If your teams spend time building filters, running reports, or finding the right module just to answer everyday ServiceNow questions, an AI assistant can take on much of that work. ServiceNow’s own Now Assist brings generative AI into the platform. Many teams also want the AI tools they already use, such as Claude and Microsoft Copilot Studio, to work with the same incident, change, and configuration management database (CMDB) data. With CData Connect AI Gateway, you set up access once and give those tools, plus LangChain and CrewAI, the same remote Model Context Protocol (MCP) URL for live ServiceNow data.

What is ServiceNow MCP and why use a managed MCP server?

ServiceNow MCP connects AI clients to incidents, CMDB items, change requests, and other data and actions. Without a shared standard, each AI tool needs its own ServiceNow connection. With MCP, you build the connection once and every tool reuses it. ServiceNow places its MCP server and client roles under Action Fabric.

Remote MCP server: An HTTP-hosted MCP endpoint that compliant AI clients reach by URL.

ServiceNow data can reach an AI client three ways:

• ServiceNow’s native MCP Server Console: It publishes Now Assist skills and platform tools. Some need Zurich Patch 9 or Australia Patch 2 or later. It suits teams using Now Assist

• A hand-built server: You write, host, and maintain it. It suits teams that want full control and have time to build

• A managed MCP platform such as CData Connect AI Gateway: The server, connectors, and request logs already exist, so you configure a connection instead of writing code. It suits teams that want to start quickly

A custom server built on the Table API leaves tool-level access controls, request logging, and maintenance to you. That is the main reason to consider a managed platform.

Before you start: prerequisites for a ServiceNow MCP server

• A ServiceNow instance URL, such as https://MyInstance12345.service-now.com/

• A dedicated integration user, not a personal account, with read access to only the objects the AI needs

• Read access to sys_db_object, sys_dictionary, and sys_glide_object through access control lists (ACLs)

• OAuth credentials (a client ID and secret) or a single sign-on (SSO) provider such as Okta, OneLogin, or PingFederate

• A CData Connect AI Gateway account. A free trial is available

• An AI client of your choice to test with

No sandbox? A free personal developer instance (PDI) from developer.servicenow.com works. Click Request Instance, then confirm the Incident list has records.

How to set up ServiceNow MCP with CData Connect AI Gateway

Step 1: Add your ServiceNow connection and authenticate

In ServiceNow:

• Open System OAuth, choose Application Registry, and click New

• Choose Create an OAuth API endpoint for external clients

• Save the client ID and secret

Then in Connect AI Gateway:

• Open Sources, click + Add Connection, and choose ServiceNow

• Enter your instance URL and select OAuth

• Add the Client Id and Client secret, click Sign in, and approve

• Click Save & Test

When Status shows Authenticated, the Data Model tab lists your ServiceNow objects. If Save & Test fails, fix the credentials and test again.

Step 2: Scope objects for incidents, CMDB, change, and request management

Permissions apply to a whole connection, not single objects, so you limit access in three places:

• ServiceNow roles and ACLs decide which objects the integration user can read

• The Permissions tab sets allowed operations. Start with Select only

• Toolkits, on plans that include them, expose only chosen objects and operations

Step 3: Get your remote MCP URL

Connect AI Gateway has one shared MCP endpoint, https://mcp.cloud.cdata.com/mcp, for any client that supports Streamable HTTP. Claude and Copilot Studio also offer built-in connectors that need no URL, while their custom options, LangChain, and CrewAI use it directly.

Add ops for a read-only URL:

https://mcp.cloud.cdata.com/mcp?ops=search_context,get_catalogs,get_schemas,get_tables,get_columns,get_instructions,queryData

For a named URL per agent or team, create a tool server:

• Open MCP Gateway, choose Tool Servers, and click + Add Tool Server

• Enter a name, select ServiceNow, and click Add

• Copy the Agent URL

Tip: give the server a clear name and, when you add it in Copilot Studio, a clear description. Copilot Studio uses the description to decide when to call the server

Step 4: Test your connection and troubleshoot common issues

Create a personal access token (PAT) on the Personal Access Tokens page, store it securely, then run:

curl -X POST https://mcp.cloud.cdata.com/mcp \
  -u ":" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","method":"tools/list","id":1}'

A tool list means the endpoint and token work. The read-only URL lists seven tools instead of 13.

• 401 errors: your PAT may have expired. Check the Security tab of Settings

• No tools listed: check your ops values. An unrecognized value returns HTTP 400

• Curl works but the client fails: check the client’s URL

To get started, connect your instance with OAuth, scope the incident, CMDB, change, and request objects, then add the MCP URL to your client.

Connect your AI tool to the ServiceNow MCP server

Claude

• In Claude, open Settings, select Connectors and click Browse connectors

• Select the CData Connect AI connector and click Connect

• Sign in to your account when prompted

To use a tool server URL or the read-only URL instead, choose Add custom connector. Enter a name and the URL, click Continue, review the detected authentication, and click Add. On Team and Enterprise plans, an Owner adds the custom connector first, and each member then clicks Connect.

Microsoft Copilot Studio

• Open your agent, go to Tools, and click Add a tool

• Search for the CData Connect AI connector and select it

• Click Create new connection and sign in

• Click Add to agent

To use a URL from Step 3 instead, choose New tool, and select Model Context Protocol. Fill in Server name, Server description, and Server URL, then pick the authentication type and sign in.

LangChain and CrewAI agents

Code clients use your email and PAT. LangChain now includes MCP support in langchain.mcp It replaces the older langchain-mcp-adapters package and its MultiServerMCPClient

Install it with

pip install "langchain[mcp]"

The module is in beta and needs version 1.4.0 or later.

from fastmcp.client import Client
from httpx import BasicAuth
from langchain.mcp import MCPAdapter

async def load_tools():
    client = Client(
        "https://mcp.cloud.cdata.com/mcp",
        auth=BasicAuth("", ""),
    )
    async with MCPAdapter(client) as adapter:
        return await adapter.list_tools()

For crewAI, pass an MCPServerHTTP entry in the agent's mcps list.

from crewai import Agent
from crewai.mcp import MCPServerHTTP

agent = Agent(
    role="ServiceNow analyst",
    goal="Answer questions about incidents and changes",
    backstory="Reads live ServiceNow data through Connect AI Gateway",
    mcps=[
        MCPServerHTTP(
            url="https://mcp.cloud.cdata.com/mcp",
            headers={"Authorization": "Basic "},
            streamable=True,
        )
    ],
)

For more detail, see the CData KB articles connecting ServiceNow to Claude, Copilot Studio, LangChain, and CrewAI.

Security and governance for enterprise ServiceNow MCP

The first security question is what happens to access control when five AI tools use the same ServiceNow instance. For ServiceNow, plan on shared authentication: every request reaches ServiceNow as the integration user, so that user’s roles and ACLs on incident, CMDB, and change data set the ceiling for every tool. If your connection shows an Authentication Model section, select Per-User Authentication, and each person signs in with their own credentials and sees only what their own ServiceNow roles allow. If it doesn’t, keep the integration user’s roles narrow.

Connect AI Gateway adds permissions that decide which operations each user or role can run on a connection and logs each request. For different access levels, create one connection per team with its own integration user, and grant each team only its own.

OAuth/Security Assertion Markup Language (SAML) passthrough: Where a source supports it, each AI user signs in with their own identity, so the AI sees only the data that user can see in the source

The MCP specification forbids a server from passing a client’s token through to downstream services and requires it to accept only tokens issued for it. The AI client signs in to Connect AI Gateway, which connects to ServiceNow with the integration user’s credentials under shared authentication, or with the user’s own credentials under per-user authentication.

To get internal approval, see the CData Get IT Approval page and ROI calculator.

Start working with live ServiceNow data using CData Connect AI Gateway

With CData Connect AI Gateway in place, adding another AI client takes only the URL you already have and a sign-in. Widen access one object at a time and test each addition with a real request. Connect AI Gateway also connects to hundreds of other enterprise sources, so the next system you add reaches your AI clients through that same URL. For prompt ideas across teams, refer CData’s prompt library.

Start your free trial today and connect your first ServiceNow instance.

Frequently asked questions