Integrations

Argus MCP server

Argus reads Splunk through MCP — and it also publishes its own MCP server, so an existing SOC copilot or any MCP host can call the full investigation workflow as tools instead of integrating a bespoke API.

Run it

bash

uv run argus mcp --transport stdio                    # for local MCP hosts
uv run argus mcp --transport streamable-http --port 8765

Hosted demo endpoint

A read-only Argus MCP server is hosted at https://mcp.tryargus.xyz/mcp so you can try the full investigation workflow with zero setup — no clone, no uv, no Splunk, no credentials. It runs against a live Splunk Enterprise + BOTS v3 and is gated by a bearer token.

bash

claude mcp add argus-hosted --transport http https://mcp.tryargus.xyz/mcp \
  --header "Authorization: Bearer <token>"

Then ask your host to use argus-hosted to investigate suspicious AWS activity in botsv3.

Read-only by design

The hosted endpoint registers only the read / investigate / proof tools below — argus_execute_response is not exposed there, so a public caller can investigate and inspect but never run containment.

Tools

Tool	Description
`argus_investigate_alert`	Investigate an alert end to end and return a grounded report. `multi_agent` and `max_turns` are optional.
`argus_recall_memory`	Recall prior cases and blocklist hits for indicators or keywords.
`argus_list_cases`	List recorded cases from KV-store case memory.
`argus_list_blocklist`	List active indicators in the threat blocklist.
`argus_list_detections`	List detections Argus has deployed as saved searches.
`argus_run_detections`	Run deployed detections on demand through MCP.
`argus_execute_response`	Execute gated response actions for a completed report.
`argus_health`	Connectivity / health check.

Response is gated, even over MCP

argus_investigate_alert is read-only. Response actions are deliberately a separate tool, argus_execute_response, and they require an explicit confirmation argument that must be exactly EXECUTE_ARGUS_RESPONSE. Anything else is refused.

json

{
  "report_json": "<the report from argus_investigate_alert>",
  "confirmation": "EXECUTE_ARGUS_RESPONSE",
  "case_id": "optional"
}

Same guarantees, different caller

Whether Argus is driven by the CLI, the dashboard, or another copilot over MCP, the rules are identical: investigation is grounded and read-only, and every write is gated.