Read-only resources
orioneye://threats/liveorioneye://campaigns/activeorioneye://telemetry/serverorioneye://intel/feed
IP usernames, server labels, and other internal fields are removed. Feed reads return published items only.
Standalone gateway · Streamable HTTP · Private preview
Connect Grok & Any LLM to Your Active Defence via MCP
Use natural language to explore OrionEye SSH telemetry, compare coordinated campaign confidence, enrich an indicator with AbuseIPDB context, and submit a finding to the moderated Intel Feed from an MCP-compatible client.
Grok/xAI directory availability subject to vendor support and approvalClient support is not implied. Claude Code supports remote HTTP MCP configuration. The current public xAI documentation does not verify a custom MCP server flow in the Grok consumer UI; connect Grok only if your client release explicitly supports remote MCP. No official xAI partnership or directory listing is claimed.
How it works
- 01 · Sign in
Authorize with OrionEye
Enterprise users can sign in with their OrionEye account or Google when configured. OAuth uses PKCE and never creates or replaces the normal web-app session.
- 02 · Connect
Add the MCP server URL
Enter the HTTPS endpoint in a client that supports MCP OAuth discovery. The gateway checks the Enterprise entitlement again for every request; Pro clients may still use an administrator-provisioned bearer where supported.
- 03 · Investigate
Ask, review, then act
Query threat intelligence or submit an item to moderation. A campaign-block request creates a proposal only; the operator must review it in OrionEye.
What the connector exposes
Threat investigation
investigate_ip(ip) combines the tracked SSH record with the AbuseIPDB dossier. It does not report or block an address.
Moderated contribution
post_intel_feed(indicator, severity) creates a submission with status in_review. A human moderator decides whether to publish it.
Action proposal
block_campaign(campaign_id) creates an HMAC-audited proposal. MCP v1 never calls Fail2Ban or the OmniAgent approval endpoint.
Bounded network tools
Enterprise clients may run ping or traceroute only for public IPv4 addresses already in OrionEye threat telemetry, and search RIPE Atlas within a radius capped at 200 km.
Telemetry scope
Health and uptime are exposed. The existing OmniAgent pairing endpoint does not provide CPU, memory, or full SRE metrics.
Example connection
For clients with MCP OAuth discovery, enter the server URL and complete OrionEye sign-in:
https://orioneye.io/mcp
For clients that require a bearer header, an administrator may provision a token securely. For Claude Code, use the remote HTTP transport:
claude mcp add --transport http orioneye \ https://orioneye.io/mcp \ --header "Authorization: Bearer YOUR_MCP_CLIENT_TOKEN"
For Claude.ai, Claude Desktop, or Grok, use the custom connector workflow only where that product, account and release support remote MCP. Provider menus and auth options can change; consult the vendor's current documentation.
Example questions:
- “Summarize the active SSH campaigns and distinguish confidence from certainty.”
- “Investigate this IP using OrionEye telemetry and its AbuseIPDB dossier.”
- “Submit this indicator to the shared feed for human moderation.”
- “Prepare a block proposal for campaign 4. Do not execute it.”
Plan access
Free
No MCP client token is issued.
Pro
Up to two clients per workspace. Read resources, threat queries and moderated Intel Feed submissions.
Enterprise
Includes proposal and bounded network tools, plus OAuth sign-in with current Enterprise entitlement checks. One gateway instance serves one workspace; OAuth tokens expire after one hour and are reissued through sign-in.
OAuth requires a client that supports MCP's OAuth discovery flow; Grok's consumer connector compatibility is not confirmed. Google sign-in appears only when configured. The gateway uses a privileged backend credential internally, never exposes it to clients, and checks the account's Enterprise plan on every OAuth-authenticated MCP request.
Security by design
- Standalone gateway, loopback binding, TLS reverse proxy, OAuth authorization code with PKCE, short-lived tokens, and fixed upstream routes.
- Passwords and Google credentials are sent only to OrionEye's backend. MCP sign-in does not create or replace the user's web-app session.
- Only token hashes are stored; revocation and rotation are administrative operations.
- Proposal and feed submission audit entries use an HMAC-linked append-only ledger. The HMAC proves ledger integrity, not an operator's digital signature or approval.
- Fail2Ban execution and AbuseIPDB report submission are not exposed because the existing backend actions are immediate and are not connected to the OmniAgent OS human-approval gate.