MCP server security for coding agents
An MCP server gets whatever your agent sends it, and what it returns goes straight to the agent. Muzzle checks both ways, and stops the agent adding servers on its own.
● Bash(claude mcp add helper -- npx helper-mcp)
⎿Muzzle blocked this: AI tools can't change their own settings. If you meant to, change it yourself outside the AI.
What Muzzle checks
Keys, tokens and your email sent in MCP callsClaude Code: Free · Codex: Pro
BlockInstructions aimed at the agent in MCP resultsPro
AlertThe agent adding a server
Blockclaude mcp add, codex mcp add, edits to .mcp.json, ~/.codex/config.toml, .cursor/mcp.json · FreeNew servers added outside the agentAlert with Undo · Pro
AlertInstruction files that tell the agent to add a serverCLAUDE.md, AGENTS.md, .cursorrules · Pro
AlertQuestions
Can I still add MCP servers?
Yes. Run claude mcp add or edit the config yourself in your own terminal. Only the agent is blocked.
Does Muzzle block an MCP server's own network traffic?
On Pro, yes for local servers Claude Code starts: their requests are checked like your scripts'. For hosted servers, Muzzle checks what the agent sends them.
Which agents?
Claude Code and Codex.
Related
Put Muzzle between your agents and the internet
Download for Mac
Free · Apple Silicon · macOS 12 or later
Pro from $8/month, or free while you contribute to the Threat Feed