Skip to content

Configuring and Operating Ranger

Once deployment is successful, the next step is understanding how to configure Ranger for your specific exercises and operate it during live runs. This section covers:

  • Connecting additional range APIs via MCP
  • Orchestrating GHOSTS through Ranger
  • Tuning prompts and logging
  • Using interfaces (chat, CLI, workflows)

Connecting Range APIs via MCP

Ranger uses the Model Context Protocol (MCP) to integrate with external services. Each service (e.g., VM orchestration, scoring) is wrapped in an MCP server.

Steps to add an MCP integration:

  1. Deploy or Enable the MCP Server:

  2. Use an existing implementation or generate one with tools like FastMCP.

  3. Base it on the service’s OpenAPI (Swagger) spec.

  4. Register the MCP Server:

  5. Set environment variables (e.g., MCP_GHOSTS_ADDR=ws://ghosts-mcp:8000).

  6. Repeat for each service (e.g., MCP_SERVICEX_ADDR=http://servicex-mcp:port).

  7. Update Tool Context:

  8. Extend Ranger’s system prompt to describe new tools.

  9. Include function names and descriptions.

  10. Test the Integration:

  11. Use a natural prompt (e.g., "create a new VM") and check if Ranger calls the MCP function.

  12. If it fails, refine tool descriptions or debug connectivity.

Design Tip: Use intuitive aliases for functions. Check open-source MCP definitions for inspiration.

GHOSTS Orchestration

Ranger simplifies GHOSTS control. Key workflows include:

  • Initial Setup: Ensure NPC agents connect to the GHOSTS server. Pre-load timeline/scripts.
  • Start NPCs: Example prompt: "Start 10 NPCs in finance segment doing office work."
  • Monitor NPCs: Ask: "How many NPCs are active and what are they doing?"
  • Adjust in Real-Time: Prompts like "Increase browsing NPCs to 20 over the next hour" are translated to API calls.

The GHOSTS MCP server supports SSE/WebSocket for live command streaming and async updates.

Prompt Management and Logging

System Prompts

  • Define Ranger’s role, available tools, and policy rules.
  • Keep it concise but authoritative.
  • Optionally include few-shot examples for behavior.

User Guidance

  • Provide a cheat sheet for phrasing requests.
  • Encourage specificity in prompts.

Logging

  • Log all interactions to a database (e.g., Baserow).

  • Review logs for:

  • Misunderstandings

  • Hallucinations
  • Performance
  • User feedback (e.g., from Open-WebUI ratings)

Prompt Adjustments

  • Refine system prompt for recurring issues.
  • Add retry/correction logic in the core.
  • Consider fine-tuning or model switching.
  • Update MCP tool schemas for clarity.

Interfaces

Web Chat (Open-WebUI)

  • Primary interface for most users.
  • Supports multi-chat, permissions, and logs for AAR.

Command-Line (CLI)

  • Use curl or scripts to call Ranger's OpenAI-compatible API:
curl http://localhost:5067/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model": "Ranger", "messages": [{"role": "user", "content": "Summarize current range status."}]}'

n8n Workflows

  • Trigger Ranger via HTTP webhooks or schedules.
  • Automate summaries, monitoring, or downstream actions.

Direct API Integration

  • Use Ranger as a backend in custom apps.
  • Follows OpenAI API structure, so most SDKs work with just a base URL change.

This configuration and operations layer turns Ranger from a chatbot into a range-integrated orchestrator. Whether you're working via chat, script, or dashboard, the same AI logic applies—making it flexible and operational in any training environment.