claude-code-mcp-manual-install
Warn
Audited by Gen Agent Trust Hub on Feb 16, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTION
Full Analysis
- [COMMAND_EXECUTION] (MEDIUM): The skill provides shell commands using
jqto directly modify~/.claude.json. This file contains the agent's session data and configuration, making it a high-value target for manipulation. - [REMOTE_CODE_EXECUTION] (MEDIUM): The proposed configuration uses
npx -y [package-name]. This facilitates the download and execution of arbitrary code from the npm registry without user interaction during the next session start. - [PERSISTENCE_MECHANISM] (MEDIUM): By modifying the
mcpServerssection of the global config file, the skill ensures that any added scripts or servers will be automatically executed every time the Claude Code environment is initialized. - [DATA_EXPOSURE] (LOW): The skill targets
~/.claude.json, which contains internal metrics and session statistics. While the instructions are for editing, an agent following these patterns could easily be redirected to exfiltrate the file content.
Audit Metadata