Find agent skills by outcome

- Ensure that despite AMBIGUITY

there is STILL A SINGLE VERIFIABLE ANSWER

- Questions may require specific output formats (datetime vs. epoch time

JSON vs. MARKDOWN)

- For example

do not count:

- File IDs

names

- URLs

GIDs

- Timestamps and datetimes (months

days

- May elicit tools returning large JSON objects or lists

overwhelming the LLM

- Use synonyms

related concepts

- Require multiple searches

analyzing multiple related items

- Complex

requiring potentially dozens of tool calls or steps

3. Questions must be REALISTIC

CLEAR

- Answerable with a single

verifiable value that you identify in advance

Create 10 human-readable questions requiring ONLY READ-ONLY

INDEPENDENT

The measure of quality of an MCP server is NOT how well or comprehensively the server implements too...

but how well these implementations (input/output schemas

- Answers must be single

verifiable values

- Questions must be READ-ONLY

INDEPENDENT

This document provides guidance on creating comprehensive evaluations for MCP servers. Evaluations t...

complex questions using only the tools provided.

- Performance testing: Check behavior under load

timeouts

- Provide helpful

specific error messages with suggested next steps

- Security testing: Validate auth

input sanitization

Writing & Content

return { content: [{ type: text

text: result }] };

Important: Annotations are hints

not security guarantees. Clients should not make security-critical decisions based solely on annotations.

- Single-user

single-session scenarios

Coding & Debugging

- Store API keys in environment variables

never in code