talocode/gatelane

GateLane is a control layer that sits between AI agents and the tools they use. When an agent wants to call a tool through the Model Context Protocol, meaning the standard way agents talk to external tools and data sources, GateLane can check who is allowed to make that call, enforce rate limits, and write down a record of what happened. Think of it as a gatekeeper that stands between your agent and every MCP server it might try to reach. Once installed through npm, you register the MCP servers you want to expose, such as a memory tool or a search tool, and GateLane discovers what tools those servers offer. From there you set allow or deny rules per tool, per server, or per individual actor, meaning a specific agent or user identity. Every call that passes through gets rate limited according to rules you define, and every result, whether allowed or blocked, is written to an audit log with its status, duration, and who made the request. GateLane ships as a command line tool with commands for managing servers, discovering and calling tools, setting policies, and reading logs, plus a local HTTP API with over twenty endpoints and a typed TypeScript client library for building the same actions into your own code. It also exposes its own MCP server, so an agent could ask GateLane itself to add a server or check a policy. Storage defaults to plain JSON files on disk, with an optional SQLite backend for persistence. Local mode needs no account or API key and, by default, allows any tool call unless you have written a rule to block it, though the project notes that production setups should define explicit allow rules instead. A separate cloud mode exists for teams that want centralized authentication using an API key, and in that mode nothing is allowed until a rule says so. GateLane is one project in a small family of open-source tools built by the same team for agent workflows.