ServiceNow as MCP Server

Overview

This skill covers setting up ServiceNow as a Model Context Protocol (MCP) server, enabling AI agents to interact with ServiceNow through a standardized protocol:

• Understanding MCP architecture and how ServiceNow fits as a tool provider
• Defining tool capabilities that expose ServiceNow operations to AI agents
• Configuring authentication and authorization for secure agent access
• Building resource endpoints that agents can discover and interact with
• Implementing input validation and output formatting for agent consumption
• Setting up logging, rate limiting, and monitoring for MCP interactions
• Testing MCP tool definitions with AI agent platforms

When to use: When enabling AI agents (Claude, ChatGPT, Cursor, etc.) to interact with ServiceNow data and operations through the MCP standard, building a ServiceNow integration for AI-powered workflows, or creating a standardized API layer for multi-agent orchestration.

Prerequisites

• Roles: admin, rest_api_explorer, web_service_admin
• Plugins: com.glide.rest (REST API), com.snc.oauth (OAuth 2.0)
• Access: Write access to sys_rest_api, sys_script_include, sys_properties, sys_oauth_entity tables
• Knowledge: REST API design, OAuth 2.0, JSON Schema, MCP specification concepts
• Related Skills: development/scripted-rest-apis for REST API development, security/acl-management for access control

Procedure

Step 1: Understand MCP Architecture

The Model Context Protocol defines a standard interface for AI agents to discover and use tools. ServiceNow acts as the MCP server, exposing capabilities that agents can invoke.

MCP ARCHITECTURE:

[AI Agent (Client)]
       |
       | MCP Protocol (JSON-RPC 2.0)
       |
[MCP Server (ServiceNow)]
       |
       +-- Tools: Operations the agent can perform
       |     +-- query_records: Search ServiceNow tables
       |     +-- create_record: Create new records
       |     +-- update_record: Modify existing records
       |     +-- run_flow: Trigger Flow Designer flows
       |
       +-- Resources: Data the agent can read
       |     +-- incident://INC0010001: Incident record
       |     +-- kb://KB0010001: Knowledge article
       |     +-- cmdb://ci/server001: Configuration item
       |
       +-- Prompts: Predefined interaction templates
             +-- incident-triage: Guide incident categorization
             +-- change-risk: Assess change request risk

Step 2: Design the Tool Schema

Define the MCP tools that expose ServiceNow operations.

Tool Definition Format (MCP Standard):

{
  "name": "sn_query_records",
  "description": "Query ServiceNow records from any table using encoded query syntax. Returns matching records with specified fields.",
  "inputSchema": {
    "type": "object",
    "properties": {
      "table": {
        "type": "string",
        "description": "ServiceNow table name (e.g., incident, change_request, sys_user)"
      },
      "query": {
        "type": "string",
        "description": "Encoded query string (e.g., priority=1^state=2)"
      },
      "fields": {
        "type": "string",
        "description": "Comma-separated list of fields to return"
      },
      "limit": {
        "type": "integer",
        "description": "Maximum records to return (default 10, max 100)",
        "default": 10
      }
    },
    "required": ["table"]
  }
}

Core Tool Definitions:

Tool Name	Purpose	Required Params	Optional Params
sn_query_records	Search records	table	query, fields, limit, order_by
sn_read_record	Get single record	table, sys_id	fields
sn_create_record	Create record	table, data
sn_update_record	Update record	table, sys_id, data
sn_get_schema	Get table schema	table
sn_run_flow	Trigger a flow	flow_id, inputs
sn_search_kb	Search knowledge base	query	kb_id, limit
sn_add_comment	Add work note/comment	table, sys_id, text	type (work_note/comment)

Step 3: Create the Scripted REST API

Build the REST API endpoints that implement MCP tool operations.

MCP Approach:

Tool: SN-Create-Record
Parameters:
  table_name: sys_rest_api
  data:
    name: "MCP Server API"
    api_id: "mcp_server"
    namespace: "x_mcp"
    short_description: "Model Context Protocol server endpoint for AI agent integration"
    active: true
    requires_authentication: true

REST Approach:

POST /api/now/table/sys_rest_api
Body: {
  "name": "MCP Server API",
  "api_id": "mcp_server",
  "namespace": "x_mcp",
  "short_description": "Model Context Protocol server endpoint for AI agent integration",
  "active": "true",
  "requires_authentication": "true"
}

Step 4: Create Resource Endpoints

Build API resources for each MCP tool capability.

MCP Approach:

Tool: SN-Create-Record
Parameters:
  table_name: sys_rest_api_resource
  data:
    rest_api: [REST_API_SYS_ID]
    name: "Tools List"
    resource_path: "/tools/list"
    http_method: "GET"
    short_description: "Return list of available MCP tools and their schemas"
    script: |
      (function process(/*RESTAPIRequest*/ request, /*RESTAPIResponse*/ response) {
        var tools = [];
        tools.push({
          name: 'sn_query_records',
          description: 'Query ServiceNow records from any table',
          inputSchema: {
            type: 'object',
            properties: {
              table: { type: 'string', description: 'Table name' },
              query: { type: 'string', description: 'Encoded query' },
              fields: { type: 'string', description: 'Fields to return' },
              limit: { type: 'integer', description: 'Max results', default: 10 }
            },
            required: ['table']
          }
        });
        // Add more tool definitions...
        response.setBody({ tools: tools });
      })(request, response);

Tool Execution Endpoint:

Tool: SN-Create-Record
Parameters:
  table_name: sys_rest_api_resource
  data:
    rest_api: [REST_API_SYS_ID]
    name: "Tools Call"
    resource_path: "/tools/call"
    http_method: "POST"
    short_description: "Execute an MCP tool with provided arguments"
    script: |
      (function process(/*RESTAPIRequest*/ request, /*RESTAPIResponse*/ response) {
        var body = request.body.data;
        var toolName = body.name;
        var args = body.arguments || {};

        var handler = new x_mcp.MCPToolHandler();
        var result = handler.execute(toolName, args);

        response.setBody({
          content: [{
            type: 'text',
            text: JSON.stringify(result)
          }]
        });
      })(request, response);

Step 5: Implement the Tool Handler

Create a Script Include to handle MCP tool execution.

MCP Approach:

Tool: SN-Create-Record
Parameters:
  table_name: sys_script_include
  data:
    name: "MCPToolHandler"
    api_name: "x_mcp.MCPToolHandler"
    script: |
      var MCPToolHandler = Class.create();
      MCPToolHandler.prototype = {
        initialize: function() {},

        execute: function(toolName, args) {
          switch(toolName) {
            case 'sn_query_records':
              return this._queryRecords(args);
            case 'sn_read_record':
              return this._readRecord(args);
            case 'sn_create_record':
              return this._createRecord(args);
            case 'sn_update_record':
              return this._updateRecord(args);
            case 'sn_get_schema':
              return this._getSchema(args);
            default:
              return { error: 'Unknown tool: ' + toolName };
          }
        },

        _queryRecords: function(args) {
          var gr = new GlideRecord(args.table);
          if (!gr.canRead()) return { error: 'Access denied to table: ' + args.table };
          if (args.query) gr.addEncodedQuery(args.query);
          gr.setLimit(Math.min(args.limit || 10, 100));
          gr.query();
          var results = [];
          var fields = args.fields ? args.fields.split(',') : null;
          while (gr.next()) {
            var record = {};
            if (fields) {
              fields.forEach(function(f) {
                record[f.trim()] = gr.getDisplayValue(f.trim());
              });
            } else {
              record.sys_id = gr.getUniqueValue();
              record.display_value = gr.getDisplayValue();
            }
            results.push(record);
          }
          return { records: results, count: results.length };
        },

        _readRecord: function(args) {
          var gr = new GlideRecord(args.table);
          if (!gr.canRead()) return { error: 'Access denied' };
          if (gr.get(args.sys_id)) {
            var record = {};
            var fields = args.fields ? args.fields.split(',') : null;
            if (fields) {
              fields.forEach(function(f) {
                record[f.trim()] = gr.getDisplayValue(f.trim());
              });
            }
            return { record: record };
          }
          return { error: 'Record not found' };
        },

        type: 'MCPToolHandler'
      };
    access: "public"
    active: true

Step 6: Configure Authentication

Set up OAuth 2.0 for secure agent authentication. Provision authentication material through an approved local credential manager or deployment secret store outside the model context. Never ask an agent to collect it, place it in a tool call, generate it in documentation, or add it to shell arguments or client JSON.

MCP Approach:

Tool: SN-Query-Table
Parameters:
  table_name: sys_oauth_entity
  query: type=client^active=true
  fields: sys_id,name,client_id,redirect_url,token_lifetime,refresh_token_lifetime
  limit: 10

Create OAuth Application:

Tool: SN-Create-Record
Parameters:
  table_name: sys_oauth_entity
  data:
    name: "MCP Agent OAuth Client"
    type: "client"
    client_id: "mcp_agent_client"
    redirect_url: "https://agent.example.com/callback"
    token_lifetime: 1800
    refresh_token_lifetime: 86400
    active: true
    comments: "OAuth client for MCP-compliant AI agents"

Authentication Flow:

1. An administrator registers the client metadata in ServiceNow.
2. An approved local credential helper performs the selected OAuth exchange outside the conversation and model context.
3. The integration transport supplies the resulting authentication state without logging it.
4. ServiceNow validates the request and applies the integration user's ACL context.

Step 7: Implement Access Controls

Configure ACLs and rate limits for MCP endpoints.

Table-Level Access Control:

Tool: SN-Query-Table
Parameters:
  table_name: sys_security_acl
  query: name=sys_rest_api^operation=execute
  fields: sys_id,name,operation,condition,role
  limit: 10

Rate Limiting Configuration:

Tool: SN-Create-Record
Parameters:
  table_name: sys_properties
  data:
    name: "x_mcp.rate_limit.per_minute"
    value: "60"
    description: "Maximum MCP API calls per minute per client"
    type: "integer"

Access Control Matrix:

Tool	Minimum Role	Table ACL Check	Rate Limit
sn_query_records	itil	Table read ACL	60/min
sn_read_record	itil	Table read ACL	60/min
sn_create_record	itil	Table create ACL	30/min
sn_update_record	itil	Table write ACL	30/min
sn_get_schema	itil	None	10/min
sn_run_flow	flow_operator	Flow ACL	10/min

Step 8: Implement Input Validation

Add validation to prevent misuse and ensure data quality.

VALIDATION RULES:

Table Name Validation:
- Must exist in sys_db_object
- Must not be in blocklist (sys_user_password, sys_certificate, etc.)
- Must be accessible to the authenticated user

Query Validation:
- Maximum query length: 2000 characters
- No script injection (block javascript: in queries for non-admin)
- Validate encoded query syntax

Field Validation:
- Fields must exist on the target table
- Sensitive fields filtered (password, token, secret fields)

Limit Validation:
- Maximum 100 records per query
- Default 10 if not specified

Data Validation (create/update):
- Validate field types match expected values
- Enforce mandatory field requirements
- Sanitize string inputs

Step 9: Set Up Logging and Monitoring

Configure logging for all MCP interactions.

MCP Approach:

Tool: SN-Create-Record
Parameters:
  table_name: sys_properties
  data:
    name: "x_mcp.logging.enabled"
    value: "true"
    description: "Enable logging for all MCP API interactions"
    type: "boolean"

Log Entry Structure:

MCP INTERACTION LOG:
- Timestamp: [ISO 8601]
- Client ID: [OAuth client]
- Tool: [tool_name]
- Arguments: [sanitized args]
- Response Status: [success/error]
- Response Time: [ms]
- Records Affected: [count]
- User Context: [authenticated user]

Step 10: Test MCP Integration

Validate the MCP server with agent platforms.

Test Tool Listing: Connect through an MCP client whose transport authentication is already configured by the local runtime, then inspect its advertised tool list.

Test Tool Execution: Invoke a low-risk, read-only schema or query tool through that MCP client. Verify allowlisting, ACL enforcement, rate limiting, sanitized logging, and response formatting. If a raw transport probe is indispensable, use an organization-approved local credential helper and keep all authentication state outside the command line and generated documentation.

Tool Usage

Tool	Purpose	When to Use
SN-Create-Record	Create REST APIs, resources, script includes	Building MCP server components
SN-Update-Record	Modify API configurations, update scripts	Iterating on MCP implementation
SN-Query-Table	Find existing APIs, ACLs, OAuth configs	Discovery and validation
SN-Get-Table-Schema	Discover table fields for tool schemas	Schema generation
SN-Get-Record	Review specific API or script details	Debugging and verification

Best Practices

1. Follow MCP specification strictly -- ensure tool schemas match the MCP JSON Schema format
2. Use OAuth 2.0 client credentials for server-to-server agent authentication
3. Implement table allowlisting -- explicitly define which tables agents can access
4. Never expose sensitive tables -- block access to credential, certificate, and password tables
5. Rate limit aggressively -- AI agents can generate high request volumes
6. Log all interactions -- maintain audit trail for security and debugging
7. Return display values -- agents work better with human-readable data
8. Implement pagination -- use cursor-based pagination for large result sets
9. Version your API -- include version in resource path for backward compatibility
10. Test with multiple agent platforms -- ensure compatibility across Claude, ChatGPT, etc.
11. Keep large reference corpora optional -- use live/source-backed lookup by default and make local indexing, SQLite, embeddings, and vector search explicit opt-ins

Troubleshooting

Issue	Cause	Resolution
401 Unauthorized	OAuth token expired or invalid	Refresh token; verify client credentials; check token lifetime
403 Forbidden	User lacks table or API ACL	Verify roles assigned to OAuth user; check table ACLs
Tool not found	Tool name mismatch between client and server	Verify tool names in tools/list response match client configuration
Empty results	Query syntax incorrect or no matching records	Test query in ServiceNow directly; check encoded query syntax
Timeout on large queries	Query returns too many records	Reduce limit; add more specific query conditions
Script error in tool handler	Runtime exception in Script Include	Check system logs (syslog); add error handling in handler scripts

Examples

Example 1: Basic Read-Only MCP Server

Scenario: Set up a minimal MCP server for read-only incident querying.

Tools exposed:

• sn_query_incidents - Search incidents
• sn_read_incident - Get incident details
• sn_search_kb - Search knowledge base

Example 2: Full ITSM MCP Server

Scenario: Complete ITSM MCP server with create/update capabilities.

Tools exposed:

• Query: incidents, changes, problems, CIs
• Create: incidents, change requests
• Update: incident state, assignment, priority
• Actions: add work notes, trigger flows
• Knowledge: search and retrieve KB articles

Example 3: Scoped MCP Server for HR

Scenario: Department-scoped MCP server for HR service delivery.

Tools exposed:

• hr_query_cases - Search HR cases (filtered to requester's own cases)
• hr_create_case - Submit new HR case
• hr_search_policies - Search HR policy knowledge base
• hr_check_benefits - Query benefits enrollment status

Access controls:

• Only hr_case table accessible
• Users can only see their own cases
• Create limited to specific case categories

Related Skills

• development/scripted-rest-apis - Building REST APIs in ServiceNow
• security/acl-management - Configuring access control lists
• admin/application-scope - Scoped application development
• genai/build-agent - Building AI agents that consume MCP tools
• genai/skill-kit-custom - Custom Now Assist skill development
• development/servicenow-docs-mcp - ServiceNowDocs search and optional local indexing for Happy Platform MCP