Query-time synthesis over the knowledge base. Retrieves the top-K grounding excerpts for a question, then asks the local model to synthesize a CITED answer grounded only in them, plus an explicit report of what the knowledge base does NOT contain (gaps). Use when you want a synthesized, sourced a...
AI agents call knowledge_think to retrieve information from Ralph Hero without modifying anything — typically the context-gathering step in research, monitoring, and reporting workflows, before the agent takes action elsewhere.
The tool performs information retrieval and synthesis over an existing knowledge base. It reads data (queries and retrieves excerpts), synthesizes answers grounded in those excerpts, and reports gaps. The description explicitly indicates it only retrieves and synthesizes information without creating, modifying, deleting, or executing external operations. This is a read-only operation with minimal security impact.
From the tool's definition Tool description states it 'retrieves the top-K grounding excerpts' and 'returns retrieved sources' — these are query operations with no data modification, deletion, or external side effects.
Attacks that exploit this kind of access
Query-time synthesis over the knowledge base. Retrieves the top-K grounding excerpts for a question, then asks the local model to synthesize a CITED answer grounded only in them, plus an explicit report of what the knowledge base does NOT contain (gaps). Use when you want a synthesized, sourced answer instead of a raw result list. Fail-open: returns retrieved sources even when the local model is offline. It is categorised as a Read tool in the Ralph Hero MCP Server, which means it retrieves data without modifying state.
Register the Ralph Hero MCP server in PolicyLayer and add a rule for knowledge_think: allow, deny, rate-limit, or require approval. Point your MCP client at the PolicyLayer proxy URL and the rule is enforced on every call, before it reaches Ralph Hero. Nothing to install.
knowledge_think is a Read tool with low risk. Read-only tools are generally safe to allow by default.
Yes. Add a rate_limit block to the knowledge_think rule in your PolicyLayer policy. For example, setting max: 10 and window: 60 limits the tool to 10 calls per minute. Rate limits are tracked per agent session and reset automatically.
Set action: deny in the PolicyLayer policy for knowledge_think. The AI agent will receive a policy violation error and cannot call the tool. You can also include a reason field to explain why the tool is blocked.
knowledge_think is provided by the Ralph Hero MCP server (ralph-hero-mcp-server). PolicyLayer sits as a proxy in front of this server to enforce policies before tool calls reach the server.
Every MCP server has a record like this.
Type a name, get the same breakdown: verified identity, auth posture, risk grade, capabilities, recommended policy.
Teams ship this data inside their own products. See what a licence covers →