The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/evaluation.py script executes a user-specified command (the MCP server) using the mcp library's stdio_client functionality. This is a fundamental component of the evaluation harness intended for testing local servers and follows the developer's provided parameters.\n- [EXTERNAL_DOWNLOADS]: The skill instructions recommend fetching official documentation and SDK README files from the modelcontextprotocol.io domain and the modelcontextprotocol organization on GitHub to ensure up-to-date protocol information.\n- [INDIRECT_PROMPT_INJECTION]: The scripts/evaluation.py harness processes external test questions from XML files and dynamically handles outputs from the MCP tools being tested, which is an inherent part of the evaluation process for LLM-based tools.\n
Ingestion points: The script reads evaluation questions from local XML files and processes responses from both the Anthropic API and the tools provided by the MCP server under test.\n
Boundary markers: The assistant is instructed via the system prompt to use specific XML-style tags (<summary>, <feedback>, <response>) to structure its output and isolate the final answer.\n
Capability inventory: The script utilizes the Anthropic Python SDK for agent interactions, performs network connections (via SSE/HTTP), and manages subprocesses (via Stdio).\n
Sanitization: Input XML is parsed using the standard xml.etree.ElementTree library, and output processing relies on regex for tag extraction.

mcp-builder