Control what your AI agents can do.

Fizzlayer sits between your agents and the tools they call. Allow what's safe, block what isn't, and approve the rest before it runs.

Open source. Self-host free with Docker.

Works withClaude CodeClaude DesktopCursorAny MCP client

Gateway

One endpoint for every tool, under your rules.

Point your agent at Fizzlayer instead of at each server. Every call goes through the rules you set.

A rule for every tool

Allow, Block or Ask first, set per tool. A tool without a rule is blocked.

Approve before it runs

A call that needs approval waits. You get a call or a webhook, approve it in the app, and the agent tries again.

Know when a tool changes

If a server rewrites a tool's description, the tool stops until you review the change.

A key for each agent

Give every client its own token. Revoke one without touching the others.

How it works

Set up in three steps.

  1. Add your servers

    Paste the URL of each MCP server. Fizzlayer lists its tools.

  2. Set the rules

    Choose Allow, Block or Ask first for each tool.

  3. Connect your agent

    Copy one config into Claude Code, Cursor or any MCP client.

Deploy

Run it your way.

Self-host

Run the gateway on your own machine with Docker. Free and open source.

Read the guide

Hosted

Create a gateway in the app and get a hosted URL for your agent. Nothing to run yourself.

Set one up

Fizzlayer Sensors

Watch the devices in your home or lab, and get a call when a reading looks wrong.

Explore Sensors

Start with one server and one rule.

The Hobbyist plan is free. Add a server, set your rules, and connect your agent.