# Navil
> Open-source runtime governance for AI agents — Cursor, Continue, Claude Code, and custom. A 2.7 µs Rust proxy in front of every MCP server your agents reach. Policy at install, enforcement at every call. Open source, Apache 2.0.

## What it is

Navil is a runtime governance proxy for AI agents. It sits between agents (Claude Code, Cursor, Continue, OpenClaw, custom) and the tools they call (MCP servers, CLIs, APIs). It enforces YAML-defined policies, scopes which tools each agent can see (94% schema-token reduction), runs 568 detection patterns across 36 agent-native attack categories, and contributes anonymized signal to a community threat network.

## Install

```
pip install navil && navil secure
```

Auto-discovers MCP configs, wraps every server, shows before/after coverage in under 60 seconds.

## Docs

- [Quickstart](https://navil.ai/docs/quickstart)
- [CLI reference](https://navil.ai/docs/cli)
- [Policy builder](https://navil.ai/docs/policy-builder)
- [Scoping (token reduction)](https://navil.ai/docs/scoping)
- [API reference](https://navil.ai/docs/api)
- [A2A agents](https://navil.ai/docs/a2a)

## Research

- [State of MCP Security (May 2026)](https://navil.ai/research/state-of-mcp-security-2026) — CVE audit of 400 MCP packages: 75% carry known vulnerabilities, Anthropic's own SDK has 6 HIGH CVEs.
- [Data download (JSON)](https://navil.ai/research/state-of-mcp-security-2026.json)

## Key pages

- [Home](https://navil.ai)
- [Live threat radar](https://navil.ai/radar)
- [MCP server registry](https://navil.ai/registry)
- [Enterprise](https://navil.ai/enterprise)
- [GitHub source](https://github.com/navilai/navil)