Sage Infrastructure Solutions Group IncSage Infrastructure Solutions Group IncLoading…

Sage Infrastructure Solutions Group Inc

AI assistants (MCP)

NullWard 0.6.108 and later have an MCP server, so an AI assistant such as Claude Code can help you configure NullWard: review WAF events and suggest exclusions, change service settings, check tunnels, and so on. The assistant connects with a short-lived personal token and works as you: it can read and change what you can in the dashboard, through the same checks, and every change is in the audit log.

Turn it on

AI assistant access is off by default. An admin turns it on under Settings > AI Assistants:

SettingWhat it does
Allow AI assistant accessOff: no token works and none can be created. Turning it off cuts every assistant off at once.
Maximum token lifetime (minutes)5 to 180 (3 hours, the most allowed). Lowering it also shortens tokens already issued.
Allow tokens that can change settingsOff: only Read tokens; existing Admin tokens become read-only.

Create a token

Under Settings > AI Assistants > Create a token:

  1. Name: what it's for, for example Claude Code: tune WAF for shop. It appears in the audit log.
  2. Access: Read only looks; Admin (can change settings) can also change settings. Changes take effect at once, so use Admin only for work you're supervising.
  3. Expires after: 15 minutes to 3 hours (up to the maximum above).
  4. Click Create token. The token is shown once: copy it, or copy the Claude Code command shown with it. Only a hash of the token is stored.

Active tokens lists your tokens with the time left and when they were last used; admins can tick All users' tokens. Revoke ends a token at once. Each person can have at most 5 active tokens.

Connect an assistant

Claude Code: run the command shown with the new token, in the project where you'll work with it:

claude mcp add --transport http nullward https://<your NullWard admin address>/api/v1/mcp \
  --header "Authorization: Bearer nw_mcp_…"

Other MCP clients: use the JSON configuration shown with the token, or set them up with:

  • URL: https://<your NullWard admin address>/api/v1/mcp
  • Transport: Streamable HTTP (JSON replies)
  • Header: Authorization: Bearer <token>

The address is Settings > Authentication > External URL if set, otherwise the address you use for the dashboard. The admin IP allowlist applies to assistants too.

Then ask the assistant, for example: "Use the nullward tools: look at the WAF events of shop and suggest exclusions for false positives." When you're done, revoke the token.

What the assistant can do

The assistant starts with an overview (you, your token, the services and tunnels) and has reference topics on services, the WAF, tunnels and the API. Its tools:

ToolWhat for
nullward_overviewYou, your token, services and tunnels at a glance
docsReference topics
list_services, get_serviceServices and their settings
update_service, create_serviceChange or add services
waf_eventsWAF events grouped by path and rule, to find false positives
list_exclusions, create_exclusion, delete_exclusionWAF exclusions (whitelist rules)
list_tunnelsApplication tunnels and their upstream groups
api_routes, api_requestThe rest of the admin API

Its instructions tell it to ask you before anything that can block or expose traffic, to change only the fields it means to, and to prefer the narrowest fix (an exclusion of the misfiring rule rather than turning the WAF off).

What it can't do

Whatever the token, these are refused (HTTP 403 with agent_token: true) and need a person in the dashboard:

  • users, their passwords and profiles;
  • AI assistant tokens and settings;
  • the dashboard's own sign-in providers, IP allowlist (admin.* settings) and certificate;
  • FIPS mode, licensing, cluster membership and diagnostics;
  • tunnel configuration downloads and install keys, which contain private keys.

It can manage sign-in providers of services, but not the dashboard's. Stored secrets (API keys, client secrets, passwords) aren't returned by the API, so the assistant doesn't see them.

Security

  • Tokens last at most 3 hours, are only accepted in the Authorization header (never in a URL), and end when they expire or are revoked, when their user is disabled or deleted, or when AI assistant access is turned off. Signing out of the dashboard doesn't end them: revoke them.
  • The assistant acts with your role and is checked like you on every request.
  • The audit log records every change as you (AI assistant token "name").
  • In a cluster, tokens work on every node, and revoking one applies everywhere.
  • Treat a token like a password while it's valid: anyone who has it can act as you.

Troubleshooting

AnswerMeaning
401 invalid, revoked or expired AI assistant tokenThe token expired, was revoked, or the maximum lifetime was lowered: create a new one.
401 AI assistant access is turned offAn admin turned Allow AI assistant access off.
403 this AI assistant token is read-onlyCreate an Admin token for changes.
403 with agent_token: trueAn agent limit: do it in the dashboard.