Partially Verified65% confidenceFactExact time
The description field in the tool metadata that an MCP server exposes to clients is essentially a natural language instruction aimed at the model, rather than documentation aimed at human users, and typically has no structured validation or input filtering.
3
Sources
65%
Confidence
Long-term
Relevance
10/14/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn MCP, tools are functions that the model can execute, resources are read-only data with no side effects, and prompts are reusable templates that users actively select.79% similarUnverifiedIn MCP tools, the docstring is passed directly to the large model as the tool description, and the model relies on this text to understand the tool's purpose and parameter meanings; therefore it must not be arbitrarily deleted.79% similarVerifiedMCP tools are essentially arbitrary code execution, indirect prompt injection is a real-world threat, and attack vectors can be hidden in resource content rather than in direct user input.77% similarUnverifiedMCP's classification standard for Tools is not based on whether they read or write, but on whether they need to execute a piece of logic based on input parameters75% similarUnverifiedMCP injects third-party text (tool names, tool descriptions, tool return results) directly into the model's context, and depending on the client implementation, this content is often assigned a higher trust weight than user input.75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1030547API
curl https://kongchang.com/api/v1/knowledge/claims/1030547MCP
get_claim(id=1030547)