Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    mcp shield

    1

    A security proxy for MCP servers that blocks destructive commands, protects sensitive files, and redacts secrets.

    Free

    1 useSecurity scanned
    mcp shield

    mcp shield

    Example session with this skill installed

    Apply the MCP security shield to my Cursor and Claude Desktop setups to prevent destructive terminal commands.

    • Read your context and instructions
    • Compiled the mcp shield

    I have initialized the security proxy for your MCP environments. This process modified your editor configuration files to route all tool calls through the protective firewall.

    Security Initialization Summary

    Target EditorActionStatusProtection Level
    Cursorwrap cursorActiveFull (Filesystem + Terminal)
    Claude Desktopwrap claudeActiveFull (Env + Tooling)
    WindsurfcheckSkippedNot Installed

    Applied Firewall Rules

    • Terminal: Blocked rm -rf, truncate, mkfs.
    • Database: Blocked DROP, TRUNCATE, GRANT.
    • Filesystem: Denied access to ~/.ssh/*, **/.env, /etc/shadow.
    • Vault: Active redaction of AWS_SECRET_ACCESS_KEY, STRIPE_SK, GITHUB_TOKEN.

    The proxy is now intercepting all JSON-RPC traffic between the LLM and your local tools. Any attempt to access restricted paths or execute destructive commands will be logged to stderr and blocked at the runtime level, returning an error to the agent without compromising the host machine.

    Next steps

    1. Restart Cursor and Claude Desktop to load the new proxy configurations.
    2. Run a test command like cat .env to verify the "Access Denied" response.
    3. Use npx mcp-shield unwrap all if you need to revert to raw execution for maintenance.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    mcp-shield is an open-source runtime security proxy and reversible secret vault for MCP (Model Context Protocol) servers that blocks destructive commands (such as rm -rf and DROP TABLE) and masks sensitive credentials before they reach language models.

    Raw execution access in MCP-enabled editors like Cursor, Claude Desktop, and Windsurf creates significant security risks. AI agents can inadvertently delete entire databases, corrupt filesystems, or exfiltrate private SSH keys through standard tool calls. This skill implements a transparent runtime security proxy that acts as a protocol-level firewall between your AI agent and its local tools.

    What it does

    • Command Filtering blocks destructive operations like rm -rf, DROP TABLE, truncate, or docker system prune before they reach the shell.
    • Path Guarding prevents the agent from reading sensitive files including .env, id_rsa, and system-level credential stores.
    • Secret Redaction & Reversible Vault replaces outbound API keys and secrets with synthetic references to prevent the LLM from seeing raw credentials.
    • Input Re-hydration maps synthetic tokens back to real secrets immediately before execution so tools continue to function seamlessly.
    • Pipe Integrity routes all security logs to stderr to ensure the stdout JSON-RPC stream remains clean and valid for the MCP host.

    How it works

    1. Use the wrap command to inject the proxy into your specific editor configuration.
    2. The agent attempts to call a tool, which is intercepted by the shield's runtime logic.
    3. The shield analyzes the command against a firewall of blocked patterns and sensitive path rules.
    4. Valid commands with redacted secrets are processed, while malicious or risky calls are terminated.

    Frameworks & tools

    This tool targets Model Context Protocol (MCP) implementations within Cursor, Claude Desktop, Claude Code, and Windsurf environments.

    Why this beats prompting it yourself

    System prompts for "safety" are easily bypassed by prompt injection or complex agent reasoning. This skill operates at the binary and protocol level, enforcing security constraints that the agent cannot see or override through conversation.

    Use cases

    • Prevent an agent from deleting production database tables or dropping data during a refactor.
    • Mask internal API keys and tokens when sending tool logs back to the LLM for analysis.
    • Block access to root-level system files and SSH configurations during autonomous coding.

    Known limitations

    Requires Node.js for the npx wrapper. Users must manually run the wrap command to initialize protection for new editors.

    Requires Node.js runtime. Needs manual re-wrapping if editor configuration files are completely overwritten by updates.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    1 use

    Times used to date

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 1 month ago

    • Free with an Agensi account

    Requires

    Shield MCP
    Node.js

    Permissions

    Terminal / Shell

    Terminal and shell access are required to wrap editor configuration files and run the security proxy process.

    Listed1 month ago

    What's inside

    Frequently Asked Questions

    MCP Shield acts as a protocol-level firewall that intercepts tool calls from your AI agent, blocking destructive commands like file deletions or database drops before they can execute. It also automatically redacts sensitive data like API keys and SSH credentials from the agent's view.

    This skill is designed specifically for environments that support the Model Context Protocol (MCP), including the Cursor editor, Claude Desktop, and Windsurf. It works by wrapping the standard tool execution process in these IDEs.

    The shield uses a 'Secret Redaction and Re-hydration' process; it replaces raw API keys with synthetic tokens when the LLM sees them, but maps them back to the original values the moment the tool executes to ensure no loss of functionality.

    Installation is straightforward: you use the provided `npx` wrapper command to inject the proxy into your specific editor's configuration file. Once wrapped, the shield runs automatically in the background whenever the editor calls a tool.

    Yes, the shield is designed to run locally and routes all security logs and alerts to `stderr`. This ensures your agent's primary communication channel (`stdout`) stays clean and prevents protocol errors during tool execution.

    New in Security