Use JSN with AI agents

Give Hermes, OpenCode, VS Code, Claude Code, Cursor, Codex, Copilot, or any shell-capable language-model harness a safe ServiceNow tool.

One CLI. Many agent harnesses.

JSN does not require a hosted MCP server. An agent uses the same local CLI a person uses, with the ServiceNow skill providing command knowledge, safety rules, and output expectations.

The integration has two parts:

Install JSN so the harness can execute jsn.
Install the ServiceNow skill so the harness knows when and how to use it.

npm install -g @jacebenson/jsn
jsn setup
jsn skill install

The default interactive installer lets you choose one or more supported targets.

Supported harnesses

Harness	Target	Install command
Hermes Agent	`hermes`	`jsn skill install --target hermes`
OpenCode	`opencode`	`jsn skill install --target opencode`
VS Code / GitHub Copilot instructions	`vscode`	`jsn skill install --target vscode`
GitHub Copilot	`copilot`	`jsn skill install --target copilot`
Claude Code	`claude`	`jsn skill install --target claude`
Cursor	`cursor`	`jsn skill install --target cursor`
Codex CLI	`codex`	`jsn skill install --target codex`
OpenClaw	`openclaw`	`jsn skill install --target openclaw`
Agents-compatible tools	`agents`	`jsn skill install --target agents`

Install to several targets at once:

jsn skill install --target hermes,opencode,claude,cursor

Or install everywhere JSN knows how to write:

jsn skill install --target all

How the agent should use JSN

The skill tells an agent to use the most specific command first, then fall back deliberately:

Use a domain command such as jsn flows, jsn catalogitems, jsn forms, or jsn logs.
Use generic records for a table that does not have a dedicated command.
Use jsn docs search for offline ServiceNow documentation.
Use jsn rest when the CLI does not expose the endpoint yet.
Treat jsn eval as a last resort.

For machine-readable results, add --json:

jsn records list --table incident --query "active=true" --json

The agent should use sys_id for mutations, identify the target profile, and ask for approval before creating, updating, deleting, or executing anything that changes an instance.

Hermes, OpenCode, Claude Code, Cursor, and Codex

These harnesses can use the installed skill as local instructions. The normal workflow is:

jsn skill install --target hermes
jsn skill install --target opencode
jsn skill install --target claude
jsn skill install --target cursor
jsn skill install --target codex

Run only the targets you use. Re-running the command refreshes the installed skill from the current JSN release.

VS Code and GitHub Copilot

Install the VS Code instruction target or the Copilot target depending on how your workspace is configured:

jsn skill install --target vscode
jsn skill install --target copilot

The CLI writes the ServiceNow instructions into the corresponding user-level instruction location. The agent still calls the local jsn executable; the instruction file is the guide, not a second API.

Any other language-model harness

A harness only needs two capabilities:

It can run a local shell command.
It can read a Markdown instruction file or accept system/tool instructions.

Install JSN for the harness user, then give the model the bundled ServiceNow skill or the repository copy:

npm install -g @jacebenson/jsn
jsn setup
jsn skill path

If the harness has a project-level instruction mechanism, copy or reference the skill at skills/servicenow/SKILL.md. Tell the agent to use jsn --help and jsn <command> --help when it needs to discover a command instead of inventing flags.

A minimal generic tool contract looks like this:

Tool: shell
Command: jsn <specific command> --json
Rules:
- Check jsn auth status first.
- Use --profile when the target instance matters.
- Prefer read-only commands.
- Ask before mutations.
- Never run logout unless explicitly requested.

That is enough to use JSN from a homegrown agent, a CI task, or another model runner without building a custom ServiceNow connector.

What JSN does not do

JSN does not give an agent permission to bypass ServiceNow ACLs, approvals, or instance policy. It does not hide mutations behind an agent abstraction. The command, profile, scope, update set, and output remain visible and inspectable.

Install JSN · Configure authentication · Use JSN safely · Agent skill feature