Tool Execution Error
A tool execution error is a failure that happens while an MCP tool runs, such as an API failure, a validation problem or a business rule the request broke. It is reported inside a successful result with isError set to true rather than in the response's error field, so anything watching status codes sees a call that worked.
Why It Matters
MCP splits failures into two kinds according to what a language model can do about them. Protocol errors cover an unknown tool, a malformed request or a server error, and they arrive in the response's error field. Tool execution errors cover a tool that ran and failed, and they arrive in the result field carrying isError. The MCP specification says clients should pass the second kind to the model, because a model can correct a bad date or an out-of-range value and try again.
A tool execution error hides better than a protocol error does. The HTTP status is 200, the JSON-RPC response is well formed and every layer built to notice failure sees success. A server author who does not trust clients to render isError can return a plain successful result describing the failure in text, which moves the failure one layer further down.
A tool execution error costs most when somebody tries to reconstruct what happened afterward. A dashboard counting HTTP errors shows nothing wrong, and the agent, which does receive the failed result, can work around it and carry on. Someone reviewing the run later sees a series of successful calls and an outcome that does not match them. Telling a failed call from a successful one means reading the result body rather than the status.