Open-source AI agent firewall that sits between agents and the network to block secret exfiltration, prompt injection, SSRF and MCP tool poisoning, with signed action receipts.
node9-ai/node9-proxy
Access control for AI coding agents: blocks credential reads, holds risky actions for review, and logs every tool call across Claude Code, Codex, Cursor and MCP servers.
- Stars
- 225
- Forks
- 21
- Open issues
- 1
- Last commit
- 9 hours ago
What is it?
node9 sits between an AI agent and the tools it calls, checking each action before it runs. Its credential jail blocks reads of paths such as ~/.ssh, ~/.aws and .env files by default, and it parses shell commands as an AST so wrapping a read does not bypass the rule. Destructive git, SQL without a WHERE clause, curl piped to bash and unauthorised sudo are held or blocked without configuration. Held actions wait for approval inside the agent conversation, and the agent is told why an action was stopped.
Beyond the defaults, node9 offers per-service rule packs for Postgres, MongoDB, Redis, AWS, Kubernetes, Docker, GitHub and the filesystem, secret and PII detection in tool arguments, an optional network egress allowlist, and an MCP gateway that pins tool definitions. It also provides a sandbox mode, a posture score for the local machine, repo scanning for CI workflows an outsider could hijack, and session history. Local enforcement needs no account; an optional login connects a machine to a shared dashboard for team policy and audit.
node9 supports Claude Code, Codex CLI, Gemini CLI, Cursor, Windsurf, VS Code, Claude Desktop and other agents, and can also run as an MCP server. It is installed with npm or Homebrew and requires Node.js 22+.
Key features
- 01Blocks agent reads of credential paths like ~/.ssh, ~/.aws and .env files.
- 02Holds destructive git, SQL and shell commands until a human approves them.
- 03Detects AWS keys, GitHub and Stripe tokens, private keys and PII in tool arguments.
- 04Wraps any MCP server and pins tool definitions so they cannot change silently.
- 05Keeps an audit record of every agent action across supported tools.
- 06Offers an optional network egress allowlist for shell commands.
Use cases
- Running Claude Code or Cursor on a laptop that holds cloud credentials and SSH keys.
- Limiting which MCP servers and tools an agent can call in a team setup.
- Reviewing risky database or shell commands before an agent executes them.
- Scanning GitHub Actions workflows for paths that could let outsiders hijack an agent.
Quick start
- 1
Install node9 globally with npm
npm install -g node9-ai - 2
Run node9 in your project to choose dashboard or local protection setup
node9 - 3
Check how exposed this machine is with the posture score
node9 posture
Things to know
- Requires Node.js 22 or newer.
- Egress control is off by default, so a command that sends a file straight to the network is not treated as a file read unless you enable it.
- Egress gating covers shell commands only, not other tools.
In collections
Topics
FAQ
- What is node9-proxy?
- Access control for AI coding agents: blocks credential reads, holds risky actions for review, and logs every tool call across Claude Code, Codex, Cursor and MCP servers. node9 sits between an AI agent and the tools it calls, checking each action before it runs. Its credential jail blocks reads of paths such as ~/.ssh, ~/.aws and .env files by default, and it parses shell commands as an AST so wrapping a read does not bypass the rule. Destructive git, SQL without a WHERE clause, curl piped to bash and unauthorised sudo are held or blocked without configuration. Held actions wait for approval inside the agent conversation, and the agent is told why an action was stopped.
- Is node9-proxy free to use? What license does it have?
- node9-proxy is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
- Is node9-proxy still maintained?
- Actively maintained. The last commit to node9-proxy was on Nov 6, 2026.
- What can I use node9-proxy for?
- Running Claude Code or Cursor on a laptop that holds cloud credentials and SSH keys. Limiting which MCP servers and tools an agent can call in a team setup. Reviewing risky database or shell commands before an agent executes them. Scanning GitHub Actions workflows for paths that could let outsiders hijack an agent.
- How do I install node9-proxy?
- Install node9 globally with npm: npm install -g node9-ai Run node9 in your project to choose dashboard or local protection setup: node9 Check how exposed this machine is with the posture score: node9 posture
- Where can I get node9-proxy?
- The source code and installation instructions are on GitHub: https://github.com/node9-ai/node9-proxy
Summary generated by AI from the project's README. GitHub stats refresh daily.
Related projects
Open-source CLI and library that verifies AI model files, code, live LLM/agent/MCP endpoints, and recorded agent traces, reporting findings and gaps.
Security scanner that checks AI agent skills (Claude Code, Codex, MCP) for prompt injection, data exfiltration and supply-chain risks before you install them.
Open standard and Python verifier that require verifiable provenance and signed evidence before an AI agent executes high-impact tool calls.
Enterprise security system that discovers, observes, benchmarks and detects risky behavior from AI coding agents and customer-facing support agents.
A security vault for OpenClaw and Claude Code that detects secrets, prompt injection and dangerous commands in AI traffic and keeps audit logs.