Estimate the token count of a string before making API calls, to stay within context limits and plan costs. e.g., text:
AI agents call estimate_tokens to retrieve information from Model Runner without modifying anything — typically the context-gathering step in research, monitoring, and reporting workflows, before the agent takes action elsewhere.
The tool only performs a read/compute operation on the provided string to estimate token usage. It does not call any external API, modify data, or trigger any side effects. It is purely informational, making it a low-severity Read category tool.
From the tool's definition "Estimate the token count of a string before making API calls" — this is a purely local computation that reads/analyzes input text and returns a count with no side effects.
Attacks that exploit this kind of access
Estimate the token count of a string before making API calls, to stay within context limits and plan costs. e.g., text:. It is categorised as a Read tool in the Model Runner MCP Server, which means it retrieves data without modifying state.
Register the Model Runner MCP server in PolicyLayer and add a rule for estimate_tokens: 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 Model Runner. Nothing to install.
estimate_tokens 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 estimate_tokens 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 estimate_tokens. 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.
estimate_tokens is provided by the Model Runner MCP server (josephtandle/replicate-mcp-mcp). 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 →