The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references and fetches documentation, README files, and SDK specifications from the official Model Context Protocol (MCP) GitHub repositories and website.
Fetches the protocol sitemap and specification from modelcontextprotocol.io.
Downloads SDK README files from the modelcontextprotocol/typescript-sdk and modelcontextprotocol/python-sdk repositories on GitHub.
[COMMAND_EXECUTION]: The provided evaluation harness (scripts/evaluation.py) supports testing local MCP servers via the stdio transport.
This functionality allows the script to launch a local process using a user-specified command and arguments (e.g., python my_server.py) to facilitate communication between the AI agent and the server under test.
[INDIRECT_PROMPT_INJECTION]: The scripts/evaluation.py script operates as an automated testing client that ingests external data in the form of evaluation questions.
Ingestion points: The script reads task questions from an XML evaluation file provided by the user via the command line.
Boundary markers: The script employs a system prompt (EVALUATION_PROMPT) that defines the agent's persona and structured response requirements (summary, feedback, response tags), though the specific test questions are interpolated directly into the message history.
Capability inventory: The evaluation loop allows the AI agent to call any tools registered by the MCP server being tested and requires interaction with the Anthropic API.
Sanitization: The script performs no specific sanitization or filtering on the questions extracted from the XML file, as it is designed for controlled testing environments.

mcp-builder