AI agents call clark_capabilities to retrieve information from Clark without modifying anything — typically the context-gathering step in research, monitoring, and reporting workflows, before the agent takes action elsewhere.
The tool queries architectural and capacity information about the Clark workforce RL agent—specifically its worker and task bounds from training. This is a retrieval operation that provides metadata without modifying state, executing code, or triggering external operations. It falls squarely within the 'Read' category as it only fetches information.
From the tool's definition Tool name 'clark_capabilities' and description 'Clark's architectural capacity: worker/task bounds it was trained' indicate it retrieves static metadata about the agent's trained limits and constraints.
Attacks that exploit this kind of access
Clark's architectural capacity: worker/task bounds it was trained. It is categorised as a Read tool in the Clark MCP Server, which means it retrieves data without modifying state.
Register the Clark MCP server in PolicyLayer and add a rule for clark_capabilities: 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 Clark. Nothing to install.
clark_capabilities 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 clark_capabilities 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 clark_capabilities. 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.
clark_capabilities is provided by the Clark MCP server (jarmstrong158/clark-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 →