Skip to content
MCP server

YUSA

Prompt-injection scanner and tool-call gate for AI agents, run locally over stdio

Runs locallyStatus and history
Pricing
Open source
Platforms
CLI, self-hosted, macOS
Runs
Locally
Version
0.3.5
Listing updated
3 days ago

About YUSA

YUSA is an open-source MCP server for security. It runs locally on your machine and is started by your AI client.

The pitch is that what a person reads on a page is not always what the agent reads. The engine decodes hidden layers such as Unicode tag smuggling, base64 blobs, HTML comments and display:none text, strips zero-width and bidi characters, and folds look-alike letters and leetspeak back to plain Latin before scoring. A directive found in a hidden layer is always a block.

Scoring uses 25 rules (instruction override, role hijack, prompt extraction, forged chat-template tokens, wallet drains, unlimited approvals, exfiltration beacons) plus 10 secret formats. Beyond text, a gate checks tool calls against a policy of allowed tools, addresses and hosts and a value cap. It also decodes raw transaction calldata, so a spender hidden in data is checked, and it blocks unlimited Permit signatures and blind eth_sign.

It is one dependency-free file with no network access, run over stdio, and installable as an .mcpb bundle for Claude Desktop. The README is explicit that the tools are advisory, since the model decides whether to call them. For a hard stop, call the library function in your own code before the tool runs. A GitHub Action can fail pull requests that add hidden instructions to files agents read.

Install

Run in a terminal

claude mcp add yusa -- npx -y https://github.com/yusasecurity/yusa/releases/download/v0.3.5/yusa.mcpb

Add to claude_desktop_config.json

{
  "mcpServers": {
    "yusa": {
      "command": "npx",
      "args": [
        "-y",
        "https://github.com/yusasecurity/yusa/releases/download/v0.3.5/yusa.mcpb"
      ]
    }
  }
}
Install in CursorOpens Cursor, which asks you to confirm. Or add it by hand:

Add to ~/.cursor/mcp.json

{
  "mcpServers": {
    "yusa": {
      "command": "npx",
      "args": [
        "-y",
        "https://github.com/yusasecurity/yusa/releases/download/v0.3.5/yusa.mcpb"
      ]
    }
  }
}
Install in VS CodeOpens VS Code, which asks you to confirm. Or add it by hand:

Add to .vscode/mcp.json

{
  "servers": {
    "yusa": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "https://github.com/yusasecurity/yusa/releases/download/v0.3.5/yusa.mcpb"
      ]
    }
  }
}

Add to ~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "yusa": {
      "command": "npx",
      "args": [
        "-y",
        "https://github.com/yusasecurity/yusa/releases/download/v0.3.5/yusa.mcpb"
      ]
    }
  }
}

Add to your client's MCP settings (most use this shape)

{
  "mcpServers": {
    "yusa": {
      "command": "npx",
      "args": [
        "-y",
        "https://github.com/yusasecurity/yusa/releases/download/v0.3.5/yusa.mcpb"
      ]
    }
  }
}

Configuration

  • YUSA_POLICYOptional path to a policy JSON (allowTools, allowAddresses, allowHosts, maxValueEth) for yusa_check_tool_call.

Tools this server provides (3)

Taken from the project documentation; the running server may differ.

yusa_scan
Scans outside text before acting on it and returns allow, review or block with findings and decoded hidden text.
yusa_fence
Wraps text in a random-nonce tag and strips invisible characters.
yusa_check_tool_call
Checks a tool call against a policy before it runs, including raw transaction calldata.

What it does

Unmask hidden text
Decodes smuggled, encoded and invisible text before scoring.
Rule-based scoring
25 injection rules and 10 secret formats yield allow, review or block.
Nonce fencing
Wraps untrusted text in a random-nonce tag so the model can tell data from orders.
Tool-call gate
Checks tools, addresses, hosts, value caps and secrets in outbound arguments.
Calldata decoding
Inspects raw transaction data for hidden spenders and recipients.
Offline engine
Runs with no dependencies and no network access.

Good for

  • Scan web pages or documents before an agent acts on them
  • Fence untrusted text before putting it in a prompt
  • Gate crypto-wallet agent calls with an allowlist policy
  • Block pull requests that hide instructions in agent config files
  • Check outbound tool arguments for leaked secrets

Pricing

Open source

Open-source repository; prices not published

Prices change. Check yusasecurity.xyz before you buy.

YUSA FAQ

Does it call any network service?

No. The engine has no network access and the server runs over stdio only.

Is it a hard guarantee?

No. The MCP tools are advisory; use the library check in your own code for a hard stop.

How do I set a policy?

Point YUSA_POLICY at a JSON file with allowTools, allowAddresses, allowHosts and maxValueEth, or pass a policy per call.

How do I install it for Claude Desktop?

Download the yusa.mcpb release file and open it.

Is YUSA free?

Yes. YUSA is open source, so you can use and self-host it for free.

How do I add YUSA to Claude Code?

Run this command in your terminal: claude mcp add yusa -- npx -y https://github.com/yusasecurity/yusa/releases/download/v0.3.5/yusa.mcpb

What are the best alternatives to YUSA?

The closest alternatives to YUSA are REA, Agent Roast, Midplane, Leash and Tinplate Edge AI Safety, Security & EU Compliance.

Alternatives to YUSA

  1. REA

    MCP server

    CLI and MCP server that lets coding agents reverse engineer apps and binaries

    Runs locally
    • Open source
    • 464K GitHub stars
    • Developer tools
  2. Agent Roast

    MCP server

    Offline review that flags agent changes which weaken tests, assertions or CI

    Runs locally
    • Open source
    • 4 Hacker News points
    • Developer tools
  3. Midplane

    MCP server

    Postgres gateway for AI agents with table grants, column masking and approval for writes

    Runs locally
    • Open source
    • 5 GitHub stars
    • Databases
  4. Leash

    MCP server

    Review what your AI touched, approve risky actions and undo supported changes

    • 1 tool
    • Remote
    • Free
    • 1 GitHub star
    • Security
  5. SecureCloud

    MCP server

    SentinelOne and IT security catalog, prices, quotes and checkout links for Israel

    • 7 tools
    • Remote
    • Free
    • Security