ActionGuard MCP server
Commit Nov 7, 2026Detects schema drift and prompt poisoning in MCP servers using cryptographic dual-hash integrity checks and a zero-trust runtime proxy.
Also known as lsendel/actionguard, actionguard
- Stars
- 0
- Tools
- —
- Downloads, 30 days
- —
- Latest version
- 1.0.0
- Last commit
- Nov 7, 2026
- License
- —
What is ActionGuard MCP?
ActionGuard detects schema drift and prompt poisoning in MCP servers using cryptographic dual-hash integrity checks and a zero-trust runtime proxy.
Summary generated from the project's README
About ActionGuard MCP server
ActionGuard Protocol Inspector Agent provides security and reliability monitoring for MCP servers. It is designed for developers and organizations running autonomous agents that rely on external tools. The system identifies schema drift and prompt poisoning by using a dual-hash integrity engine that creates separate cryptographic fingerprints for JSON Schemas and tool descriptions. This ensures that silent mutations in tool definitions are caught before they affect agent behavior.
Users can implement this via a zero-trust runtime proxy that routes tool calls through a Cloudflare-based edge firewall. When a tool deviates from its pinned hash, the proxy terminates the request with an HTTP 423 Locked error. The service is accessible through a CLI, as a GitHub Actions CI gate, or as a native MCP server using stdio or streamable HTTP modes. It requires a target URL and a pinned hash to maintain security boundaries during execution.
What you can do with it
- 1Prevent breaking schema changesIntegrate a verification step into your CI pipeline to block unannounced tool mutations from reaching production.
- 2Secure agent tool executionRoute your agent tool calls through an edge firewall to stop drifted or poisoned payloads from reaching your LLM.
- 3Audit MCP endpoint securityRun a live protocol inspection against any MCP or A2A endpoint to receive a reliability scorecard.
Written from the project's README. Every use case is checked against a quote from it.
How to install ActionGuard MCP
Formats verified 10 Nov 2026Add to CursorAdd to GooseInstall ActionGuard MCP in Claude Code
$ claude mcp add --transport http actionguard https://actionguard.ai/mcpInstall ActionGuard MCP in Cursor
~/.cursor/mcp.json
{
"mcpServers": {
"actionguard": {
"url": "https://actionguard.ai/mcp"
}
}
}Install ActionGuard MCP in VS Code
.vscode/mcp.json
{
"servers": {
"actionguard": {
"type": "http",
"url": "https://actionguard.ai/mcp"
}
}
}Install ActionGuard MCP in VS Code CLI
$ code --add-mcp '{"name":"actionguard","type":"http","url":"https://actionguard.ai/mcp"}'Install ActionGuard MCP in Codex
$ codex mcp add actionguard --url https://actionguard.ai/mcpInstall ActionGuard MCP in Codex config.toml
~/.codex/config.toml
[mcp_servers.actionguard] url = "https://actionguard.ai/mcp"
Install ActionGuard MCP in Gemini CLI
$ gemini mcp add -s user --transport http actionguard https://actionguard.ai/mcpInstall ActionGuard MCP in Gemini settings.json
~/.gemini/settings.json
{
"mcpServers": {
"actionguard": {
"url": "https://actionguard.ai/mcp",
"type": "http"
}
}
}Automated checks
Ran Nov 15, 2026Full report →Not verified yet. A server needs a score of 80, no serious finding, and enough data for us to check. How checks work · What to fix for more points
- Source repository linked and reachablelsendel/actionguard is public and reachablegithub.com/lsendel/actionguard ↗Pass
- Package names this repository as its sourceNot checked
- Build provenance (Sigstore attestation)Not checked
- Open source license (OSI)No licenseFinding
- Commit or release in the last yearLast commit or release 2026-11-07github.com/lsendel/actionguard ↗Pass
- OpenSSF ScorecardNot checked
- Package name not close to a popular packageNot checked
- No critical advisory in the package or its dependenciesNot checked
- No hidden instructions to the model0 tool descriptions and the README read; no override phrases, credential paths or hidden charactersPass
- Destructive tools are labelledNot checked
- Remote endpoint answers the MCP handshakeinitialize answered with protocol 2024-11-05 in 140 msPass
README
View on GitHubActionGuard.ai
Drift and poisoning detection for MCP servers.
"SSL Labs for MCP servers. Know when your tools change, not just when they go down."
Diff detection for MCP tools: catch silent schema drift and poisoned tool descriptions before your autonomous agents run them.
Live Production Services
- 🌐 Web Inspector & Diagnostic Engine:
https://actionguard.ai/ - 📊 Public Reliability Index:
https://actionguard.ai/index - 🎮 Interactive Drift & Prompt Poisoning Simulator:
https://actionguard.ai/playground - ⭐ Self-Serve Domain Ownership Portal:
https://actionguard.ai/verify - 💳 Developer API Keys & Pricing:
https://actionguard.ai/pricing - 📖 Protocol Guides & Specs:
https://actionguard.ai/docs/mcp-schema-drift
Quickstart
1. GitHub Actions CI Gate
Block breaking changes and unannounced schema mutations before they hit production:
name: CI
on: [pull_request]
jobs:
verify-agent-tools:
runs-on: ubuntu-latest
steps:
- name: Verify Upstream MCP Tools
uses: lsendel/actionguard@v1
with:
target_url: 'https://github.mcp.modelcontextprotocol.io/mcp'
pinned_hash: '81d36f5215b689d6f100cfe77096fbe8d113c304c0dd8374d69a84678fb69415'
2. Zero-Trust Runtime Gateway Proxy ("Cloudflare for MCP")
Route agent tool calls through ActionGuard's edge firewall to terminate drifted tools with HTTP 423 Locked:
{
"mcpServers": {
"weather": {
"command": "npx",
"args": [
"-y",
"@modelcontextprotocol/server-weather",
"--proxy",
"https://actionguard.ai/proxy/v1?target=https%3A%2F%2Fapi.weather.com%2Fmcp&pin=a1b2c3d4..."
]
}
}
}
3. Embed Live Status Badges in Your Repository
Show real-time verification and protocol compliance on your MCP server's GitHub repository.
Markdown (Native SVG)
[](https://actionguard.ai/servers/<your-domain-or-slug>)
Shields.io Endpoint
[](https://actionguard.ai/servers/<your-domain-or-slug>)
- Domain Verified Badge: Gold (
#f59e0b) badge⭐ A • Domain Ownerdisplayed when domain ownership is claimed via DNS TXT or.well-known/actionguard-verification. - Reliability Grades: Dynamic color coding based on cryptographic dual-hash baseline and 5-stage protocol inspection (
A+,A,B,C,D,F).
Claude Desktop & Cursor (claude_desktop_config.json)
{
"mcpServers": {
"github": {
"url": "https://actionguard.ai/v1/proxy?target=https://github.mcp.modelcontextprotocol.io/mcp&pin=81d36f5215b689d6f100cfe77096fbe8d113c304c0dd8374d69a84678fb69415"
}
}
}
Python (LangChain / CrewAI / httpx)
import httpx
GATEWAY = "https://actionguard.ai/v1/proxy"
TARGET = "https://github.mcp.modelcontextprotocol.io/mcp"
PIN = "81d36f5215b689d6f100cfe77096fbe8d113c304c0dd8374d69a84678fb69415"
# Requests fail with 423 Locked if upstream schema drifts or tool is poisoned
client = httpx.Client(base_url=f"{GATEWAY}?target={TARGET}&pin={PIN}")
3. Run as an MCP Server (Claude Desktop, Cursor, Glama)
You can run ActionGuard natively as a tool inside any MCP client:
Stdio Mode (Claude Desktop claude_desktop_config.json)
{
"mcpServers": {
"actionguard": {
"command": "npx",
"args": ["-y", "@actionguard/cli", "mcp"]
}
}
}
Streamable HTTP Mode
{
"mcpServers": {
"actionguard": {
"url": "https://actionguard.ai/mcp"
}
}
}
4. CLI Quick Run
# Run 5-stage live protocol audit against any MCP or A2A endpoint
npx @actionguard/cli inspect https://api.myservice.com/mcp
# Compile single actionguard.json to MCP server.json, A2A agent-card.json, and DNS-AID records
npx @actionguard/cli compile ./actionguard.json
# Local CI check
npx @actionguard/cli check https://api.myservice.com/mcp --pin <hash>
Core Product Pillars
- The 5-Stage Live Inspector ("SSL Labs for Agents"): Zero-signup diagnostic tool running live edge probes against MCP and A2A endpoints, returning an A+ to F scorecard and instant copy-paste fixes.
- Dual-Hash Integrity Engine: Distinct cryptographic fingerprints for JSON Schemas (
schema_sha256) versus tool descriptions (descriptions_sha256) to catch silent prompt mutations. - Enterprise Dependency Watchlist: Automated 15-minute edge monitoring with instant multi-channel alerts (Slack Block Kit, Discord Embeds, Email via Resend, and signed Webhooks).
- Self-Serve Domain Proof (DNS-AID): Cryptographic domain verification via DNS TXT records or well-known challenge files to earn the Gold "⭐ Domain Verified" badge.
- Zero-Trust Runtime Firewall: Low-latency proxy (<15ms) that blocks execution before poisoned or drifted payloads reach your LLM.
Infrastructure Architecture
Built entirely on the Cloudflare edge stack for global sub-50ms latency:
- Cloudflare Workers: Edge probing engine, public API, zero-trust proxy, and dynamic SVG badges.
- Cloudflare D1: High-performance relational edge database for accounts, catalog, subscriptions, and API keys.
- Cloudflare R2: Content-addressed, immutable storage for historical snapshot payloads, raw schemas, and audit logs.
- Continuous Monitoring: 15-minute cron trigger with automatic drift detection and Resend email / Slack dispatch.
Monorepo Packages
├── packages/
│ ├── core/ # @actionguard/core — Dual-hash, linter, compiler, and probes
│ └── cli/ # @actionguard/cli — actionguard command line tool
└── apps/
└── worker/ # Cloudflare Worker — Web Inspector, API, proxy, and D1/R2 storage
Documentation
- Product Requirements Document (PRD)
- Protocol Specifications & Mappings
- System Architecture & Data Model
- Step-by-Step Implementation Plan
- Hacker News Show HN Submission
- Product Hunt & Twitter Launch Kit
License
MIT © ActionGuard.ai
ActionGuard MCP questions
How do I add ActionGuard MCP to Claude Code?
Run claude mcp add --transport http actionguard https://actionguard.ai/mcp in your terminal. Add --scope user to make it available in every project.
How do I add ActionGuard MCP to Cursor?
Open ~/.cursor/mcp.json and add the server under mcpServers, using the Cursor config shown in the install section above.
What does ActionGuard detect?
It identifies silent schema drift and poisoned tool descriptions in MCP servers to prevent autonomous agents from running compromised tools.
How does the zero-trust runtime gateway work?
It acts as a low-latency proxy that terminates drifted tools with an HTTP 423 Locked status before they reach your LLM.
Can I use ActionGuard in my CI/CD pipeline?
Yes, you can use a GitHub Actions CI gate to block breaking changes and unannounced schema mutations before they reach production.
What is the Dual-Hash Integrity Engine?
It uses distinct cryptographic fingerprints for JSON Schemas and tool descriptions to catch silent prompt mutations.