Claude Auto-Approve MCP

Claude Auto‑Approve MCP

The Claude Auto‑Approve MCP solves a common friction point for developers who use Claude Desktop in automated or highly interactive workflows: the need to manually approve every tool invocation. When an AI assistant requests a tool, Claude Desktop presents a dialog asking the user to allow or block that request. For rapid iteration or production pipelines this manual step becomes a bottleneck, especially when the same tools are repeatedly used across multiple chats or projects. The Auto‑Approve MCP restores a safer, more granular approval model that was unintentionally disabled by an update introducing an “Always approve” button. By configuring permanent approvals and per‑chat decisions, developers can keep the security of explicit prompts while eliminating repetitive clicks.

At its core, the server restarts Claude Desktop with a debugger port enabled and injects JavaScript that hooks into Claude’s internal tool‑approval mechanism. The injected code automatically grants or blocks tools based on rules defined in the MCP configuration ( and ). Because it operates inside the running instance, approvals are applied instantly without restarting the assistant for each new chat. The server also offers a option on macOS that installs a lightweight watcher and LaunchAgent, ensuring the debugger port is always active from login onward. This persistence layer removes the need for manual restarts after system reboots, giving developers a truly hands‑free environment.

Key capabilities include:

• Granular control: Specify tool names to auto‑approve or block per MCP server, allowing different projects to have distinct policies.
• Per‑chat flexibility: The injected logic can change decisions for new chats while preserving previous approvals, striking a balance between automation and oversight.
• Automatic environment setup: The server handles the restart of Claude Desktop with the debugger port, eliminating manual configuration steps.
• Persistence on macOS: Optional watcher scripts and LaunchAgents keep the debugger port alive across reboots, reducing downtime.

Typical use cases involve continuous integration pipelines where a Claude instance must run scripted tests or generate code, data‑science notebooks that repeatedly call external APIs, and collaborative environments where multiple developers share a single Claude Desktop instance but need isolated tool access. In all scenarios, the MCP removes the manual approval barrier while still allowing developers to audit and adjust permissions as needed.

By integrating seamlessly with existing MCP workflows—declaring the server in and passing simple JSON lists of tools—the Auto‑Approve MCP offers a lightweight, secure, and developer‑friendly solution to one of the most annoying aspects of using Claude Desktop for automated tasks.