ServiceNow MCP Server

ServiceNow's MCP Server is a platform-native implementation of the Model Context Protocol that lets external AI agents (Claude, Copilot, Vertex AI, Bedrock, or custom agents) execute governed enterprise actions on a ServiceNow instance. Unlike data-only integrations, it exposes the full "system of action": flows, playbooks, approvals, catalog items, and Now Assist skills across IT, HR, customer service, security, risk and compliance, and app development.

Every invocation runs through the ServiceNow AI Control Tower (AICT), so calls are identity-verified, permission-scoped, metered, and fully auditable. The MCP Server Console provides automatic tool discovery, managed OAuth, session management, role-based tool packages, consumption metering, and enterprise audit trails. AI Gateway adds discovery, approval workflows, pause controls, and PII-based sensitive data protection at the gateway layer.

The server is generally available and included in every Now Assist and AI Native SKU (shipped with Zurich Patch 4 / Yokohama Patch 11). It is provider-hosted: there is no repo to clone. Administrators provision an MCP server inside their ServiceNow instance, wrap Now Assist skills as MCP tools, and external clients connect to a per-instance endpoint over Streamable HTTP carrying JSON-RPC 2.0, authenticated via OAuth 2.1.

Tools

ServiceNow's MCP Server is provider-hosted on your ServiceNow instance. There is no package to install on the client side beyond pointing your MCP client at the instance endpoint.

Prerequisites

• ServiceNow instance on Zurich Patch 4 or Yokohama Patch 11 (or later)
• Now Assist or AI Native SKU entitlement (included)
• Admin access to the MCP Server Console and Machine Identity Console
• Confirm /sncapps/mcp-server routing on the instance (test that https://<instance>.service-now.com/sncapps/mcp-server/mcp/sn_mcp_server_default returns HTTP 401, not 404; open a HI case if missing)

Provision the server (Instance B, tool side)

1. Configure an OAuth 2.1 inbound integration in the Machine Identity Console; capture Client ID and Client Secret.
2. Author the Now Assist skills you want to expose.
3. In the MCP Server Console, create an MCP server and wrap each skill as an MCP tool.
4. Assign role-based tool packages and activate the server.

Connect a client (Instance A or external agent) Endpoint format:

https://<instance>.service-now.com/sncapps/mcp-server/mcp/<server-name>

Example MCP client config (Claude Desktop / Cursor style for remote MCP):

{
  "mcpServers": {
    "servicenow": {
      "url": "https://<instance>.service-now.com/sncapps/mcp-server/mcp/<server-name>",
      "transport": "http",
      "auth": {
        "type": "oauth2",
        "authorization_url": "https://<instance>.service-now.com/oauth_auth.do",
        "token_url": "https://<instance>.service-now.com/oauth_token.do",
        "client_id": "<client_id>",
        "client_secret": "<client_secret>",
        "redirect_uri": "https://<your-redirect>/oauth_redirect.do"
      }
    }
  }
}

For a ServiceNow-to-ServiceNow setup, on the agent-side instance create a Connection & Credential Alias, register Instance B's server in the MCP Client (AI Agent Studio), then bind the discovered tools to your AI Agent. API Key and Connection & Credential Alias auth modes are also supported.

Governance All traffic flows through AI Control Tower. Use AI Gateway to discover, approve, pause, and monitor servers; enable Sensitive Data Protection to block PII at the gateway.

• Let an external Claude or Copilot agent open, update, and resolve ServiceNow incidents on behalf of an end user with full audit trails in AICT.
• Expose HR Service Delivery skills (onboarding requests, case creation, knowledge lookup) as governed tools for an enterprise chat agent.
• Allow a security operations agent to trigger SecOps playbooks, enrich indicators, and create response tasks without bypassing ServiceNow controls.
• Drive Customer Service Management workflows from a contact-center copilot: create cases, route approvals, and pull catalog items.
• Build cross-instance multi-agent systems where one ServiceNow agent invokes governed tools on another ServiceNow instance over MCP.

• "Create a P2 incident in ServiceNow for the payments outage and assign it to the SRE on-call group."
• "Find my open HR cases and trigger the offboarding playbook for employee ID 48211."
• "Submit a catalog request for a new MacBook Pro for Priya Singh and route it for manager approval."
• "Pull the last 10 SecOps response tasks tagged ransomware and summarize their status."
• "Approve change request CHG0045123 and add a note that the CAB review is complete."

• Official, platform-native server maintained by ServiceNow and included in every Now Assist and AI Native SKU at no extra license cost.
• Enterprise-grade governance: AICT identity verification, OAuth 2.1, role-based tool packages, consumption metering, and full audit trails.
• Spans the entire ServiceNow estate (ITSM, HR, CSM, SecOps, GRC, app dev) rather than just one product.
• AI Gateway adds discovery, approval workflows, pause controls, and PII protection across all MCP traffic.

• Only remote MCP servers are supported; local stdio servers are not available.
• Requires a ServiceNow instance on Zurich Patch 4 / Yokohama Patch 11 or later, plus admin work to wrap skills as tools and configure OAuth.
• Each tool invocation consumes Now Assist assists (standard skill cost plus one additional assist), so high-volume agent traffic has a metered cost.
• GCC environments are not yet supported for the MCP Server (MCP client and A2A are).

• michaelbuckner/servicenow-mcp: community open-source MCP server that wraps ServiceNow REST APIs directly, useful when you cannot use the platform-native server.
• echelon-ai-labs/servicenow-mcp: another community implementation focused on natural-language access to ServiceNow records.
• Direct ServiceNow Table API or Flow Designer integrations via a generic HTTP MCP server, for teams that prefer raw REST.