Agents & MCP
AI in nano: the built-in pivt assistant, and two MCP servers that let coding agents author detections and run SOC operations
Choose an AI workflow to author detections or investigate your data. Use pivt inside the product, which is included with nano Self-run (paid), or connect a coding agent through the MCP servers on any edition.
nano has two kinds of AI help:
- pivt, the assistant inside the product. It drafts and tunes detections, generates dashboards and parsers, writes nPL, investigates new cases and triages alerts. It's included with nano Self-run (paid), which has every feature at every volume, and stays off until you configure your own OpenAI-compatible or Anthropic endpoint. The free open-core edition doesn't include it.
- MCP servers that let a coding agent (Claude Code, Codex, Cursor, Antigravity, or any MCP client) drive nano from your terminal or CI. They work with nano Self-run and also with the free open-core edition (AGPL-3.0), where they're the way to get AI-assisted authoring and investigation.
The MCP servers
- nanodac runs detection-as-code. An agent drafts detections, validates and tests them, and ships them to nano through pull requests. It writes rules.
- nano-investigator runs SOC operations. An agent searches, triages alerts, works cases and notebooks, hunts, builds dashboards and authors parsers. It reads first, and writes cases, notebooks, dashboards and parsers.
Both run locally over stdio, read NANOSIEM_API_URL and NANOSIEM_API_KEY, and can run side by side.
| nanodac | nano-investigator | |
|---|---|---|
| Job | Author and ship detections | Search, triage, investigate, hunt, build parsers |
| Writes | Detection rules | Cases, notebooks, dashboards, parsers |
| Workflow | GitOps: draft, validate, PR, sync | Interactive: ask, follow the investigation |
| Surface | CLI, 26 MCP tools, GitHub Action | 91 MCP tools, 5 prompts |
| Package | @nano-rs/dac-cli, @nano-rs/dac-mcp-server | @nano-rs/investigator-mcp-server |
They meet in one place: when a hunt in nano-investigator finds something that deserves a standing rule, author that rule with nanodac.
Both can produce parsers. Detection-as-code drafts a version-controlled parser.yaml and validates it in the repository; nano-investigator authors and publishes a parser interactively against your deployment.
Connect to your deployment
Point either server at your nano base URL (the app host your users open, for example https://nano.example.com), not the separate ingest hostname.
Create the API key under Settings → Access Control → API Keys. A key carries an explicit permission list and can't exceed its owner's permissions. The Tool Reference lists the permission each tool needs.
To deploy nano and onboard log sources with the same coding agent, see Set up nano with your coding agent.
Agent Lane and agent skills
- Agent Lane (nano Self-run) gives AI agents a read-only, audited fast lane onto your event data in native ClickHouse SQL, with their own named, expiring tokens.
- Agent skills are official skills for coding agents:
nano-deploydeploys and operates nano Self-run, andnano-agentlanehunts through Agent Lane. See Hunt with your AI agent.
Sections
pivt
The built-in assistant: @ commands, AI wizards and background investigation.
Agent Lane
A read-only, audited fast lane for AI agents onto event data (Self-run).
Agent skills
Official nano-deploy and nano-agentlane skills for coding agents.
Detection-as-Code
Author parsers, detections and searches with a coding agent and nanodac.
Investigator
Drive nano as a SOC analyst workstation over MCP: triage, investigate, hunt.
Troubleshoot
Need help with these checks? Contact support with the request ID and time with timezone.
An agent uses UDM field names
The investigator prompts and resources in version 0.1.16 still use UDM names. Give the agent the OCSF field reference and inspect the query before running it.
What's next
Audit and troubleshooting
Find what SOAR did and why: audit events, run statuses, failure reasons, and fixes for flows that don't fire, runs that fail, and dropped trigger events.
pivt
nano's built-in AI assistant: the @ command panel, AI wizards for detections, tuning, dashboards, parsers and queries, and background case investigation.