- Post History
- Subscribe to RSS Feed
- Mark as New
- Mark as Read
- Bookmark
- Subscribe
- Printer Friendly Page
- Report Inappropriate Content
10m ago
Connect any MCP client to ServiceNow Otto for Customer Service Management case tools
CSM MCP Server exposes ServiceNow Otto for Customer Service Management's case tools to any Model Context Protocol (MCP) client, such as Moveworks or Claude in the web. A connected client can look up cases and case tasks, summarize a case, draft resolution notes, check sentiment, and draft an activity response, all through natural-language requests instead of the CSM UI. Every request passes through the same roles and permissions a human agent already works under, so a connected AI client never returns more case data than the person asking is allowed to see.
CSM MCP Server is delivered as its own app (sn_csm_mcp_server), built on the Model Context Protocol Server app (sn_mcp_server) and the ServiceNow Otto for Customer Service Management plugin (sn_csm_gen_ai). None of the three activates automatically with another; each must be installed and active on its own before CSM MCP Server's tools can be used. The four GenAI skill tools (case summarization, resolution notes generation, sentiment analysis, activity response) additionally require the com.sn.csm.gen.ai plugin and only install where that plugin is active.
Key Capabilities:
- Case and case task lookup: get_customer_service_cases and get_case_tasks let a connected client retrieve case and case task records directly, scoped to the roles assigned to that connection.
- Case summarization: case_summarization returns a live summary of a case's activity and current status without opening the CSM UI.
- Resolution notes generation: resolution_notes_generation drafts resolution notes from a case's own work notes and activity.
- Sentiment analysis: sentiment_analysis returns a live sentiment score, trend, and reasoning pulled from the case record.
- AI-drafted activity response: activity_response drafts a follow-up message in the case's own voice for a live agent to review and send.
- Permission-scoped by design: Every tool call is checked against the connecting user's existing CSM roles before any case data is returned, so a connection never returns more than that role already permits.
- Single activation toggle for GenAI tools: case_summarization, resolution_notes_generation, sentiment_analysis, and activity_response are tied to one toggle, so they turn on together only for customers with the underlying Otto for CSM AI skills active.
Implementation Steps
From All > MCP Server Console > Configuration > Servers, select CSM MCP Server and switch the application scope to CSM MCP Server. Set the Auth scope field to a2aauthscope to enable the REST A2A protocol for the AI Agent MCP tools, then set the Deactivate/Activate toggle to Activate. This creates an OAuth client entry (for example, sn_csm_mcp_server.csm_default); set up OAuth using the auto-populated Authorization code grant fields, or your own OAuth client if you configure one directly with the oauth_admin or admin role. Activating requires the sn_mcp_server.admin or admin role.
The Get Case Tasks tool needs two additional manual steps beyond this before it works: enable subflow tools under the MCP Staging Subflow Tools category (off by default), then manually activate the underlying subflow and the Get Case Tasks tool itself.
Before pointing any of the four GenAI skill tools (case summarization, resolution notes generation, sentiment analysis, activity response) at real customers, clone the underlying Otto for CSM skill rather than turning on the out-of-the-box version directly. Ground the clone in your own case data and access rules, permitted actions, case relationships, and prompt wording, run evals against that grounded version, and only then publish it for real end users. The out-of-the-box skill is a starting example, not a finished skill to activate as-is.
- Activate the CSM MCP Server
- CSM MCP Server (names admin, sn_customerservice_agent, sn_customerservice_consumer_agent, sn_csm_mcp_invoke; does not mention snc_internal — see Best Practices note below)
- CSM MCP Server tools reference (also names sn_mcp_server.viewer for base tool access; this page writes the consumer-agent role with a dot, sn_customerservice.consumer_agent — standardized on the main CSM MCP Server page's no-dot spelling, sn_customerservice_consumer_agent, throughout this article per OPM direction)
- Set up OAuth for CSM MCP Server
- Configure a client to connect to an MCP server
- Exploring the MCP Server Console
Community Resources
Best Practices
- Assign roles per tool, not all at once: Grant sn_csm_mcp_invoke, sn_mcp_server.viewer (base tool access), and sn_customerservice_agent or sn_customerservice_consumer_agent deliberately per tool, rather than granting access to all four GenAI skill tools by default. [OPM to confirm: snc_internal for Get Case Tasks is not named in either public CSM MCP Server docs page checked — confirm via BT1/internal source before publishing, or remove if it was a stale carryover.]
- Treat the four GenAI tools as one switch: Confirm all four (case summarization, resolution notes generation, sentiment analysis, activity response) are appropriate for a customer before activating, since the single toggle turns them on together.
- Plan for the Get Case Tasks manual steps: Budget time for the subflow activation steps above. Automated builds cannot currently activate flows for this tool.
- Account for per-call billing on external calls: A call to a CSM MCP Server tool from outside ServiceNow is billed one ServiceNow Otto Assist per call. A call from a tool inside ServiceNow is not billed an Assist.
- Set approval behavior on the connecting client: Whether a tool call runs with or without a person approving each step first is configured on the connecting AI client (for example, in AI Agent Studio), not inside CSM MCP Server itself.
Formal Learning
Measured Success & Outcomes
| Outcome | How It Supports Teams | Key Metric |
|---|---|---|
| Faster, more consistent case resolutions | AI clients call CSM's case tools directly, with no custom integration required | Case resolution time from a connected AI client |
| Case data opened to outside AI clients without new access controls | Every request still passes through CSM's existing roles and permissions | Unauthorized-access incidents from connected AI clients (target: zero) |
| Faster case prioritization and resolution | AI and human agents get real-time sentiment scoring and drafted activity responses | Time from case sentiment flag to first response |
FAQs
Q. What do I need before I can activate CSM MCP Server?
Three plugins must be active: ServiceNow Otto for Customer Service Management (sn_csm_gen_ai), Model Context Protocol Server (sn_mcp_server), and CSM MCP Server (sn_csm_mcp_server). Activating the connection requires the sn_mcp_server.admin or admin role.
Q. Does a connected AI client see the same case data a human agent would?
Yes. Every CSM MCP Server request is checked against the connecting user's existing CSM roles and permissions before any case data is returned. A connection is bound to those roles, not to a separate, wider grant.
Q. Why does the Get Case Tasks tool need extra setup?
Get Case Tasks depends on a subflow, and automated builds cannot currently activate flows for this tool. Enable subflow tools under the MCP Staging Subflow Tools category, then manually activate the underlying subflow and the tool itself.
Q. Am I billed differently for calls that come from outside ServiceNow?
Yes. A call to a CSM MCP Server tool from outside ServiceNow is billed one ServiceNow Otto Assist per call. A call from a tool inside ServiceNow is not billed an Assist.
Q. Can I turn on the GenAI skill tools individually?
No. case_summarization, resolution_notes_generation, sentiment_analysis, and activity_response share a single activation toggle, and all four only install where the com.sn.csm.gen.ai plugin is active.
Q. Should I activate the out-of-the-box GenAI skill tools directly for my customers?
No. Clone the underlying Otto for CSM skill, ground it in your own case data, access rules, and prompt wording, and run evals against that grounded version before publishing it for real end users. Treat the out-of-the-box skill as a starting example, not a finished skill to activate as-is.
Q. Which ServiceNow package includes CSM MCP Server?
CSM MCP Server is included in CSM Advanced. Under the legacy SKU naming, that is Pro Plus or Enterprise Plus.
Q. Does CSM MCP Server support the MCP standard's prompts and resources capabilities?
No. CSM MCP Server supports case lookup, summarization, and action-style tools only. Prompts and resources are not supported, a deliberate security decision applied across the whole MCP platform to reduce prompt-injection risk.
Ready to connect your AI client?
Back to Implementation Library Home