armguard-mcp
Safety-enforcing MCP server for ROS 2 robot arms with approval prompts and audit log
- Pricing
- Open source
- Platforms
- CLI, Linux, self-hosted
- Runs
- Locally
- Version
- 0.1.1
- Listing updated
- today
Similar MCP servers
About armguard-mcp
armguard-mcp is an open-source MCP server for engineering simulation. It runs locally on your machine and is started by your AI client.
The key claim is that limits live on the server, not in the prompt. Every plan is checked against a YAML policy before anything moves: joint limits, a workspace box, keep-out zones, speed and step caps, force thresholds, allowlists and rate limits, with a software e-stop on top. A prompt injection hidden in an image or topic string can change what the model asks for but not what the server allows.
Motion is two-step. plan tools return a preview and a single-use plan_id, and only execute_plan moves the arm. Motion outside the envelope needs a human to approve it through MCP elicitation, and is denied by default when the client cannot ask. All calls, approvals, denials and e-stops go into an append-only JSONL audit log.
It is alpha. The 22 tools run against a simulated Franka FR3, and the ROS 2 backend with MoveIt 2 was tested on Jazzy but not on a physical robot. The README states it is defense in depth, not certified functional safety.
Install
Run in a terminal
claude mcp add armguard-mcp -- uvx armguard-mcpAdd to claude_desktop_config.json
{
"mcpServers": {
"armguard-mcp": {
"command": "uvx",
"args": [
"armguard-mcp"
]
}
}
}Add to ~/.cursor/mcp.json
{
"mcpServers": {
"armguard-mcp": {
"command": "uvx",
"args": [
"armguard-mcp"
]
}
}
}Add to .vscode/mcp.json
{
"servers": {
"armguard-mcp": {
"type": "stdio",
"command": "uvx",
"args": [
"armguard-mcp"
]
}
}
}Add to ~/.codeium/windsurf/mcp_config.json
{
"mcpServers": {
"armguard-mcp": {
"command": "uvx",
"args": [
"armguard-mcp"
]
}
}
}Add to your client's MCP settings (most use this shape)
{
"mcpServers": {
"armguard-mcp": {
"command": "uvx",
"args": [
"armguard-mcp"
]
}
}
}Using several servers? Build one config file with the MCP config generator.
What it does
- Server-side envelope
- Joint limits, workspace box, keep-out zones, speed, step and force caps.
- Plan then execute
- Previews return single-use plan ids; only execute_plan moves.
- Human approval
- Out-of-envelope motion asks via MCP elicitation.
- Audit log
- Append-only JSONL of calls, approvals and e-stops.
- Simulated demo
- A bundled FR3 policy runs with one uvx command.
- Narrow scope
- 22 fixed tools instead of arbitrary ROS access.
Good for
- Try an LLM controlling a simulated robot arm safely
- Test prompt-injection resistance
- Preview a motion before executing
- Keep an audit trail of agent-driven robot actions
Pricing
Open source
Free open-source package
armguard-mcp FAQ
Has it run on a real robot?
No, the README says it has not been run on a physical robot yet.
Is it certified safety?
No, it is defense in depth, not certified functional safety.
Does it need ROS to try?
No, the simulated FR3 starts with uvx armguard-mcp.
How does it differ from generic ROS bridges?
It exposes 22 fixed skill-level tools and blocks arbitrary publish or service calls.
Is armguard-mcp free?
Yes. armguard-mcp is open source, so you can use and self-host it for free.
How do I add armguard-mcp to Claude Code?
Run this command in your terminal: claude mcp add armguard-mcp -- uvx armguard-mcp
What are the best alternatives to armguard-mcp?
The closest alternatives to armguard-mcp are REA, Context7, Chrome DevTools MCP, Codebase Memory and GitHub.
Where this comes from
- Description: written from the project's README.
- Install steps: from the official MCP Registry entry.


