The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill directs the agent to fetch documentation and SDK README files from official Model Context Protocol (MCP) sources.
Official documentation: https://modelcontextprotocol.io/sitemap.xml
SDK documentation: https://raw.githubusercontent.com/modelcontextprotocol/typescript-sdk/main/README.md and https://raw.githubusercontent.com/modelcontextprotocol/python-sdk/main/README.md
[COMMAND_EXECUTION]: The evaluation.py script is designed to launch MCP servers using local shell commands as part of a developer's testing workflow.
Evidence: The script accepts -c (command) and -a (args) parameters to initiate server processes for evaluation.
[DYNAMIC_EXECUTION]: The connections.py utility supports the dynamic execution of local server binaries to establish communication via the standard input/output (stdio) transport.
Evidence: The MCPConnectionStdio class uses the mcp library's stdio_client to execute user-specified commands.
[INDIRECT_PROMPT_INJECTION]: The testing framework ingests data from external XML files and tool outputs, which constitutes a potential surface for indirect prompt injection.
Evidence: The agent_loop in scripts/evaluation.py processes responses from external MCP tools and user-provided test questions, though this is a standard risk for agent-based testing tools.

mcp-builder