multi-repo-context-MCP-server

The server dynamically loads tool modules using `import()` based on paths specified in its configuration files (e.g., `repo.tools` property). If a malicious actor can compromise or inject into the `.multi-repo-mcp/repos.json` (project-local or user config) or control the `MULTI_REPO_MCP_CONFIG` environment variable, they could specify paths to arbitrary code, leading to Remote Code Execution (RCE). Additionally, the `search_across_repos` tool uses a user-provided `query` in a regular expression, which could be vulnerable to Regular Expression Denial of Service (ReDoS) attacks. File system operations (`fs.readFile`, `fs.readdir`) are used, which, while generally restricted to configured `repoPath`s, could pose a risk if `repoPath` validation is circumvented (e.g., via symlinks, though `RepoValidator` attempts to prevent this).