Coding Tools MCP

Coding Tools MCP turns a chat client such as Claude Desktop into a coding agent on one repository. It reads and searches files, applies multi-file patches, runs commands (including interactive ones under a real terminal), and shows git status, diffs, logs and blame. Everything is confined to the workspace folder you start it with: absolute paths, `..` and symlink escapes are refused, and on Linux, Landlock adds kernel-level filesystem confinement. The default `safe` mode asks for permission before network-looking commands, shell expansion, inline scripts and destructive commands; `trusted` opens those, and `dangerous` is meant only for containers or VMs. `exec_command` still runs real shell commands in the workspace, so point it at code you'd let an agent change. The server sends anonymous usage telemetry (tool names and counters) to PostHog by default; set `CODING_TOOLS_MCP_TELEMETRY=off` or `DO_NOT_TRACK=1` to turn it off.

Configuration

{
  "mcpServers": {
    "coding-tools": {
      "command": "uvx",
      "args": ["coding-tools-mcp", "--stdio", "--workspace", "/path/to/repo"]
    }
  }
}

Add this to your claude_desktop_config.json file. Copied verbatim from the upstream README.