Run Agent-Assisted Research

​

Requirements

• A connected MCP client with the FullHunt server loaded
• A valid API key
• Account access for any Enterprise, Nexus, or OEM tools in the workflow

​

Step 1: verify access

Call auth_status. Return only the authentication result, plan,
remaining credits, and enabled module fields. Stop after this call.

​

Step 2: search

Call search with query="CVE-2024-3094". Return each result id, title,
and type exactly as supplied by FullHunt. Do not fetch a result yet.

​

Step 3: fetch one result

Select the vulnerability result from the previous FullHunt search.
Call fetch with its exact id. Return:

- fields observed in the fetched response
- missing fields needed for remediation
- source identifiers and timestamps
- analysis in a separate section

Do not call active scan tools.

​

Step 4: make evidence-driven pivots

​

Required output contract

{
  "question": "",
  "tool_calls": [],
  "observations": [],
  "analysis": [],
  "missing_evidence": [],
  "recommended_next_query": null
}

​

Safety and repeatability

• Specify passive tools when active scanning is not required.
• Require confirmation before fullhunt_scan, enterprise_on_demand_scans, or oem_on_demand_scan.
• Limit tool calls and pagination per run.
• Store results, identifiers, and timestamps outside the conversation.
• Require the model to identify uncertainty and absent fields.
• Re-run the same prompt against a fixed test input after tool or model changes.