cd /news/ai-agents/argos-a-1mb-rust-security-shim-for-m… · home › topics › ai-agents › article
[ARTICLE · art-146814] src=github.com ↗ pub= topic=ai-agents verified=true sentiment=↑ positive

Argos –- A 1MB Rust Security Shim for Model Context Protocol Servers

A developer released Argos, a ~1 MB single-binary Rust security shim that sits between AI clients such as Claude Desktop and Cursor and Model Context Protocol (MCP) tool servers, inspecting raw JSON-RPC traffic to block unauthorized file access, path traversal, and destructive commands. Argos enforces workspace boundaries via OS path canonicalization and '../' stripping, shields files including .env, id_rsa, id_ed25519, and cloud credentials, intercepts commands such as 'rm -rf', 'mkfs', and fork bombs, and writes blocked and allowed actions to a local JSON-lines log (argos-audit.log) with no cloud telemetry, configured through argos.toml. The project claims sub-millisecond overhead, with the inspection of tools/call shown at under 0.2ms, and ships pre-compiled binaries for Windows and Linux plus source builds via cargo build --release.

read3 min views2 publishedOct 7, 2026
Argos –- A 1MB Rust Security Shim for Model Context Protocol Servers
Image: Michielbdejong (auto-discovered)

Zero-overhead policy enforcement gateway and runtime guardrail for Model Context Protocol (MCP) servers.

argos acts as a transparent security pipe between AI clients (Claude Desktop, Cursor) and underlying MCP tool servers. Inspired by the zero-friction philosophy of Quad9/Pi-hole, it inspects raw JSON-RPC traffic on the fly and deterministically blocks unauthorized file access, path traversal attacks, and destructive commands before they reach your system.

  • Sub-millisecond Overhead: Built with pure Rust and Tokio asynchronous streams. Zero perceptible lag for the agent or developer.
  • Path Traversal Sandboxing: Enforces strict workspace boundaries via OS path canonicalization and../ stripping.
  • Secret & Sensitive File Shield: Block access to.env , private keys (id_rsa ,id_ed25519 ), and cloud credentials.
  • Destructive Command Blocker: Intercepts dangerous terminal commands (rm -rf , disk formatters, fork bombs).
  • Local Audit Logging: Records blocked and allowed actions into a structured, JSON-lines log (argos-audit.log ) without cloud telemetry.
  • Flexible Configuration: Declarative rule customization viaargos.toml .
  • Your Own Local & Private Tool : Built entirely in Rust as a self-contained, ~1 MB single binary with zero external telemetry or cloud dependencies. Argos relies strictly on deterministic pattern matching, native OS primitives, and JSON-RPC stream interception—ensuring your sensitive code, configuration keys, and audit trails never leave your local machine.

Download the latest pre-compiled binary for your system from the Releases page:

  • Windows : Downloadargos.exe (or unpackargos-windows-x86_64.zip ).
  • Linux : Download and extractargos-linux-x86_64.tar.gz :
tar -xvf argos-linux-x86_64.tar.gz
chmod +x argos

If you are running macOS (Apple Silicon / Intel) or prefer compiling locally:

git clone [https://github.com/JUSICK/Argos-mcp-guardrail.git](https://github.com/JUSICK/Argos-mcp-guardrail.git)
cd Argos-mcp-guardrail
cargo build --release

The compiled binary will be located at:

  • Linux / macOS: target/release/argos
  • Windows: target/release/argos.exe

Create and place argos.toml next to the argos executable or in your workspace root:

[filesystem]
block_path_traversal = true

blocked_patterns = [".env", ".ssh", "id_rsa", "id_ed25519", "credentials", ".aws", ".npmrc"]

allowed_patterns = [".env.example", ".env.sample", ".env.template"]

[commands]
blocked_commands = [
  "rm -rf",
  "mkfs",
  ":(){ :|:& };:",
  "chmod -R 777",
  "dd if="
]

[audit]
enabled = true
log_allowed = false
log_file = "argos-audit.log"

Update your claude_desktop_config.json:

{
  "mcpServers": {
    "Argos": {
      "command": "C:\\path\\to\\argos.exe",
      "args": [
        "--",
        "npx.cmd",
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "C:\\Users\\username\\projects\\my-workspace"
      ]
    }
  }
}
{
  "mcpServers": {
    "Argos": {
      "command": "/usr/local/bin/argos",
      "args": [
        "--",
        "npx",
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "/Users/username/projects/my-workspace"
      ]
    }
  }
}

Restart Claude Desktop, and Argos will actively guard your tool calls against unauthorized filesystem traversal and credential exposure.

[ AI Client (Claude / Cursor) ]
              │
              │ stdin / stdout (JSON-RPC)
              ▼
   ┌───────────────────────┐
   │         argos         │  <── Inspects tools/call in <0.2ms
   └───────────────────────┘
         │           │
   (If Allowed)  (If Blocked) ──> Returns JSON-RPC Error & logs event
         │
         ▼
[ Real MCP Tool Server ]

You are able to have as many Argos as you want, change their names e.g. "Argos-Backend", "Argos-Frontend" for a big project that has 2 or more AI agents.

MIT License. Free for personal and commercial use.

── more in #ai-agents 4 stories · sorted by recency
── more on @argos 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/argos-a-1mb-rust-sec…] indexed:0 read:3min 2026-10-07 · —