AI agents call get_credential to retrieve information from Authbox without modifying anything — typically the context-gathering step in research, monitoring, and reporting workflows, before the agent takes action elsewhere.
While reading credentials is nominally a Read operation (no side effects on the credential data itself), the severity is elevated to 'high' because: (1) credentials are sensitive assets whose disclosure enables unauthorized access to downstream systems; (2) the server implements policy-gating and scope controls suggesting recognition of this risk; (3) an AI agent with unconstrained access to this tool could leak or…
From the tool's definition Tool description explicitly states 'Retrieve a credential' with 'Returns credential fields', indicating data retrieval. No modification, deletion, or execution described.
Attacks that exploit this kind of access
Retrieve a credential from the Auth Box vault. Returns credential fields filtered by the agent\. It is categorised as a Read tool in the Authbox MCP Server, which means it retrieves data without modifying state.
Register the Authbox MCP server in PolicyLayer and add a rule for get_credential: 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 Authbox. Nothing to install.
get_credential 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 get_credential 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 get_credential. 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.
get_credential is provided by the Authbox MCP server (marucie/authbox). 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 →