AI agents call ai.describe-image to retrieve information from Mcp without modifying anything — typically the context-gathering step in research, monitoring, and reporting workflows, before the agent takes action elsewhere.
This tool processes an image input and returns descriptive metadata (caption and structured details) via Claude Haiku vision. It is a read-only analysis operation analogous to 'get' or 'fetch' — it retrieves information from the image without creating, modifying, deleting, or executing anything.
From the tool's definition Tool description states it 'Describe[s] an image' and 'Returns caption + structured details' — a pure retrieval and analysis operation with no side effects, data modification, code execution, or external state changes.
Attacks that exploit this kind of access
Describe an image (JPEG/PNG/GIF/WebP, ≤1MB) via Claude Haiku vision. Returns caption + structured details. It is categorised as a Read tool in the Mcp MCP Server, which means it retrieves data without modifying state.
Register the MCP server in PolicyLayer and add a rule for ai.describe-image: 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 Mcp. Nothing to install.
ai.describe-image 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 ai.describe-image 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 ai.describe-image. 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.
ai.describe-image is provided by the MCP server (@2sio/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 →