AI agents call benchmarks.get to retrieve information from Or Info without modifying anything — typically the context-gathering step in research, monitoring, and reporting workflows, before the agent takes action elsewhere.
This tool retrieves existing benchmark data (ELO rankings) from LMArena without modifying, deleting, or executing operations. It is a read-only query operation with no capability to alter system state or trigger external actions. The blast radius of misuse is minimal—an agent could only retrieve information already publicly available.
From the tool's definition Tool description states 'Get LMArena ELO ranking for a model' with retrieval of 'score, global rank, vote count and confidence interval'. The verb 'Get' and the nature of queried data (rankings, scores) indicate pure data retrieval with no side effects.
Attacks that exploit this kind of access
Get LMArena ELO ranking for a model: score, global rank, vote count and confidence interval. It is categorised as a Read tool in the Or Info MCP Server, which means it retrieves data without modifying state.
Register the Or Info MCP server in PolicyLayer and add a rule for benchmarks.get: 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 Or Info. Nothing to install.
benchmarks.get 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 benchmarks.get 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 benchmarks.get. 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.
benchmarks.get is provided by the Or Info MCP server (jmtrs/or-info). 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 →