The full upstream README, mirrored here for reference. Install config, tool schemas, adoption signals, and an original overview live on the ServiceNow MCP Server listing page.
The developer data plane for ServiceNow in Claude Code — Table API, schema, aggregates and update sets,
on any release from Tokyo up, on PDIs, in GCC, for free. Runs alongside ServiceNow's native MCP Server.
Table API · CMDB · Update Sets · Aggregation · Resources · Read-only mode · Audit log · OAuth 2.1+PKCE · Streamable HTTP · Claude Code Plugin · 4 Skills
This MCP server gives AI assistants the raw ServiceNow data plane: any table via the Table API, the data dictionary (with inherited fields), Stats API aggregates, CMDB classes and relationships, and update sets. 19 tools and 5 resources, with tool annotations, an optional read-only mode, table allow/deny lists, hardened encoded queries, and a JSON audit line per call.
Built with FastMCP 4.0 — speaks both the stateless MCP 2026-07-28 protocol and the legacy handshake, negotiated per connection.
ServiceNow's MCP Server Console (Action Fabric) exposes Now Assist skills, Knowledge Graph, flows, scripted REST and playbooks as governed tools — and excludes the Table API by design ("cannot be converted to MCP tools regardless of configuration"). This project is the other half: the developer loop that native does not serve. ServiceNow's own CEG AI CoE guide lists uvx mcp-server-servicenow as its community "Path C".
| Native MCP Server Console | This project | |
|---|---|---|
| Tool sources | Now Assist skills, Knowledge Graph, subflows/actions, scripted REST (GET/POST/PUT), playbooks, MCP Apps | Table API CRUD, Stats API aggregates, sys_dictionary schema, update sets, CMDB via Table API |
| Table API | Excluded by design | Any table, any field |
| Minimum release | Zurich P9 / Australia P2 for custom tools | Tokyo+ |
| Entitlement | Now Assist / AI-Native SKU (docs: Prime for inbound); metered in assists | MIT, $0, your own compute |
| PDIs / GCC | Not available | Works |
| Auth | OAuth 2.0 auth-code via Machine Identity Console (JWT, no DCR) | OAuth 2.1 + PKCE proxy (DCR + CIMD), static tokens, or a service account |
| Governance | AI Control Tower / AI Gateway | Self-managed: --read-only, table allow/deny, opt-in write confirmation, audit log, tool annotations |
| Transport | Streamable HTTP only | stdio + Streamable HTTP |
| Resources / prompts | Roadmap | 5 resources |
Run both. Native tells Claude what ServiceNow means (skills, summaries, semantic search); this server tells Claude what ServiceNow contains (rows, schema, aggregates, update sets). No tool-name collisions today.
Sign up for a free Personal Developer Instance (PDI) — it comes pre-loaded with demo data. Wake it from the developer portal if it's hibernating.
Note: Instances with ServiceNow's basic-auth restriction enforced (the default on new PDIs since mid-2026) reject REST basic auth with
401 "Required to provide Auth information"unless the integration user has thesnc_basic_auth_api_accessrole. Grant it via User Administration → Users → your user → Roles.
Copy .mcp.json.example to .mcp.json and fill in your credentials, or use the Claude Code CLI:
Ask Claude: "List the 5 most recent incidents" — if it returns data, you're connected.
| Tool | Description |
|---|---|
list_records | List records from any table with filtering, field selection, and pagination |
get_record | Get a single record by sys_id |
create_record | Create a new record in any table |
update_record | Update an existing record (optional confirmation, see below) |
delete_record | Delete a record by sys_id (optional confirmation, see below) |
aggregate_records | COUNT, AVG, MIN, MAX, SUM with GROUP BY + HAVING via Stats API |
| Tool | Description |
|---|---|
list_ci | List configuration items with class and query filtering |
get_ci | Get a single CI by sys_id |
create_ci | Create a new configuration item |
update_ci | Update a configuration item (optional confirmation, see below) |
get_ci_relationships | Get parent/child relationships for a CI (paged: limit, offset) |
| Tool | Description |
|---|---|
get_system_properties | Query system properties |
get_current_user | Get authenticated user info |
get_table_schema | Table data dictionary incl. inherited fields (hierarchy, per-field defined_in) |
| Tool | Description |
|---|---|
list_update_sets | List update sets with state filtering |
get_update_set | Get update set details |
create_update_set | Create a new update set |
set_current_update_set | Set the active update set |
list_update_set_changes | List changes within an update set |
MCP Resources provide read-only context that LLM clients can fetch without tool calls — reducing latency and token overhead.
| Resource URI | Description |
|---|---|
servicenow://schema/{table_name} | Field definitions (name, type, label, mandatory, reference, defined_in) for any table, parents included |
servicenow://instance | Instance URL, platform version, logged-in user, timezone |
servicenow://update-set/current | Currently active update set name, sys_id, state |
servicenow://cmdb/classes | CMDB CI class hierarchy (names, labels, parent classes) |
servicenow://help/query-syntax | Encoded query operators reference (prevents hallucinated syntax) |
Every tool carries MCP annotations (readOnlyHint, destructiveHint, idempotentHint, openWorldHint) so clients such as Claude Code can apply their own permission policy. On top of that, the server ships its own guardrails — all off by default except the audit log:
| Flag | Env var | Effect |
|---|---|---|
--read-only | SERVICENOW_READ_ONLY=true | Hides the 7 write tools from tools/list and refuses any non-GET request at the HTTP layer |
--table-allowlist a,b* | SERVICENOW_TABLE_ALLOWLIST | Only these tables (exact or glob) may be accessed — applies to table_name, class_name, fixed-table tools and the schema resource |
--table-denylist sys_user* | SERVICENOW_TABLE_DENYLIST | These tables may never be accessed; wins over the allowlist |
| (always on) | — | Table names, sys_ids, field lists and order_by are validated; javascript: in caller queries is limited to gs.* date helpers (gs.daysAgo(7), gs.beginningOfToday() …) |
--allow-js-queries | SERVICENOW_ALLOW_JS_QUERIES=true | Re-enable arbitrary javascript: in queries |
--write-confirm | SERVICENOW_WRITE_CONFIRM=true | Ask the user before update_record, update_ci, delete_record (see below) |
--audit-log stderr|off|PATH | SERVICENOW_AUDIT_LOG | One JSON line per tool call / resource read (default: stderr) |
An audit line (keys only — never values):
With --write-confirm, the three destructive tools fetch the record's current values and ask the user before writing:
elicitation/create). Decline stops the write; the assistant is told not to retry.InputRequiredResult; the server never sends one to a client that has not declared elicitation.confirmation_required error carrying the preview; the assistant shows it and re-runs with confirm=true.It is off by default because the MCP client's own permission prompt is the primary human-in-the-loop, and because automation cannot answer a dialog. Enable it where an AI Steward wants a second gate.
Add to your MCP client config — copy the snippet for your tool:
Add to ~/Library/Application Support/Claude/claude_desktop_config.json:
Add to .cursor/mcp.json or .vscode/mcp.json:
See Configuration Guide for OAuth, multi-instance, and the full environment variable reference.
See Deployment Guide — Docker, Cloud Run, HTTP transport verification, and the security model.
See TROUBLESHOOTING.md for common issues (hibernating instances, 401 errors, OAuth).
This project ships 4 Claude Code skills in skills/ (installed via the plugin, not the PyPI package) — guided workflows that chain MCP tools for common ServiceNow tasks. Skills auto-trigger from natural conversation or can be invoked directly.
| Skill | What It Does | Try Saying |
|---|---|---|
| servicenow-cmdb | CI classes, dependencies, CMDB health, data quality, CSDM compliance | "show me CMDB health" / "what depends on this server" |
| exploring-tables | Schema discovery, field types, data profiling, table comparison | "what fields does incident have" / "find tables matching cmdb" |
| reviewing-update-sets | Update set review, risk flagging, conflict detection, pre-promotion checks | "review my update sets" / "is this safe to promote" |
| triaging-incidents | Incident triage, priority assessment, CI correlation, bulk analysis | "what's on fire" / "open P1 incidents" |
The update set reviewer is a unique differentiator — no other open-source ServiceNow MCP server provides guided update set review workflows with risk categorization and pre-promotion checklists.
Install as a Claude Code plugin for zero-config setup — the MCP server, skills, slash commands, and admin agent are bundled together.
Set these environment variables (or add them to your shell profile):
| Command | Description |
|---|---|
/servicenow:triage | Triage incidents — list, investigate, assess priority, analyze trends |
/servicenow:cmdb | Explore CMDB — CI hierarchy, dependencies, health, CSDM taxonomy |
/servicenow:review-update-set | Review update sets — deep review, compare, pre-promotion checks |
/servicenow:explore-table | Explore tables — schema, fields, data profiling, table search |
The servicenow-admin agent handles complex multi-step tasks autonomously (CMDB audits, incident trend reports, batch update set reviews). Claude can spawn it as a background worker for long-running analysis.
Note: The plugin auto-configures the MCP server — no manual
.mcp.jsonsetup required.
aggregate_records Stats API tool