Hitl Mcp Server — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited Hitl Mcp Server (Agent Skill) and scored it 45/100 (orange). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 1 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 1 flagged
A base64 string of 128+ characters appears in a documentation file. Encoded prompt injection hides the hostile instruction in base64 — invisible to keyword filters — and relies on the agent's ability to decode it at runtime. There is no normal authoring reason to embed a multi-hundred-byte base64 blob in skill docs.
*.sig, SIGNATURES) outside the documentation.Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
1) Sign up for AgentMP.ai and get an API Key:
https://agentmp.ai/createapikey
(You will have to sign in using your gmail/linkedIn)
2) Get Local Setup Docker Image and Run it: (Prereq - Docker desktop installed & running)
docker pull agentmp/hitl-mcp-server:latest
docker run -d \ --name hitl-mcp-server \ --env AGENTMP_API_KEY="<YOUR AGENTMP API KEY>" \ --restart unless-stopped \ -i \ agentmp/hitl-mcp-server
3) Configure claude_desktop.json
{ "mcpServers": { "hitl": { "command": "docker", "args": ["exec", "-i", "hitl-mcp-server", "node", "app.js"], "env": { "AGENTMP_API_KEY": "your_api_key_here" } } } }
4) Use Cogito HITL on Claude
eg.
List all my HITL escalations
Create a HITL escalation with these details:
Get details for HITL escalation "esc-[ID-FROM-TEST-2]"
Approve HITL escalation "esc-[ID-FROM-TEST-2]" with comment "Approved for testing purposes"
Create a HITL escalation for a $50 software license purchase, then modify it to $75 enterprise license
5) To receive Cogito push notifications on Slack/Teams/WhatsApp/Telegram , go to https://agentmp.ai/linkchannels , select your preferred channel and configure your channel identifier.
For Teams and Slack contact your channel admin to enable AgentMP bot.
We are actively onboarding other channels. If you need to run this MCP server entirely within your firewall reachout to [email protected]
6) To test out Cogito HITL API on postman you can use postman collection - HITL.postman_collection.json - in this repo.
Please refer to this demo video:
https://github.com/AgentMP/hitl-mcp-server/blob/main/HITLDemoVideo.mp4
HITL MCP server streamlines user interactions with AI Agents that are semi-autonomous. Agents can perform their tasks and reach out to humans for approvals, status updates, feedbacks, course corrections using this MCP tool. User interactions are captured, stored for future enhancements. User interactions are managed in channels of choice by end user (slack/teams/whatsapp/telegram/web browser).
There does not seem to be a single standard for HITL.
A Model Context Protocol (MCP) server that bridges Claude Desktop with your HITL (Human In The Loop) escalation service. This allows Claude to create, approve, modify, reject, and manage HITL escalations directly from Claude Desktop.
https://hub.docker.com/r/agentmp/hitl-mcp-server
(Please refer to SetupClaudeDesktop.md for steps)
mkdir hitl-mcp-server
cd hitl-mcp-server export AGENTMP_API_KEY="your_api_key_here" chmod +x run.sh
./run.shOr manually:
docker-compose up --build -d npm install export AGENTMP_API_KEY="your_api_key_here" chmod +x setup-local.sh
./setup-local.sh node app.jsBefore configuring Claude Desktop, test your setup:
# Test the server directly
export AGENTMP_API_KEY="your_api_key"
node app.jsIf working correctly, you should see:
🚀 Starting HITL MCP Server...
✅ API Key loaded successfully
🔧 Setting up request handlers...
✅ HITL MCP Server initialized successfully
🌟 HITL MCP Server running and connected via stdio
📡 Ready to handle HITL requests from Claude DesktopTo use this MCP server with Claude Desktop, you need to configure it in your Claude Desktop settings.
#### For Docker Deployment:
{
"mcpServers": {
"hitl": {
"command": "docker",
"args": ["exec", "-i", "hitl-mcp-server", "node", "app.js"],
"env": {
"AGENTMP_API_KEY": "your_api_key_here"
}
}
}
}#### For Local Node.js Deployment:
{
"mcpServers": {
"hitl": {
"command": "node",
"args": ["/path/to/your/hitl-mcp-server/app.js"],
"env": {
"AGENTMP_API_KEY": "your_api_key_here"
}
}
}
}Once configured, you can use the following commands in Claude Desktop:
Create a HITL escalation for session "sess-123" with agent "my-agent" for a purchase of "Premium License" costing $99.99. Escalate to user "[email protected]" with normal priority.Approve HITL escalation "esc-123456789" with comment "Approved for purchase"Modify HITL escalation "esc-123456789" to change the item to "Enterprise License" and amount to $199.99 with comment "Upgraded to enterprise"Reject HITL escalation "esc-123456789" with comment "Budget exceeded"Get details for HITL escalation "esc-123456789"List all my HITL escalationsThe MCP server provides the following tools:
AGENTMP_API_KEY (required): Your API key for the AGENTMP platformThe server communicates with your HITL service using these endpoints:
POST /api/hitl/mcp - MCP JSON-RPC endpoint for create, approve, modifyPOST /api/hitl/{id}/reject - REST endpoint for rejecting escalationsGET /api/hitl/{id} - REST endpoint for getting escalation detailsGET /api/hitl - REST endpoint for listing escalationsdocker-compose up -d to start itdocker ps | grep hitl-mcp-server to verify it's runningnpm install in the project directoryAGENTMP_API_KEY environment variable is properly setdocker logs hitl-mcp-serversetup-local.shbackend.agentmp.ioFor Docker deployment, view logs with:
docker logs hitl-mcp-serverFor local deployment, logs will appear in the terminal where you ran npm start.
Feel free to submit issues and enhancement requests!
MIT License
export AGENTMP_API_KEY="99aeae0e6b4f4107a59d1b59010834cb3eb96ae0f667078f7cc2a296dad27785"
docker build -t hitl-mcp-server .
docker run -d \ --name hitl-mcp-server \ --env AGENTMP_API_KEY="99aeae0e6b4f4107a59d1b59010834cb3eb96ae0f667078f7cc2a296dad27785" \ --restart unless-stopped \ -i \ hitl-mcp-server
Perfect! Your Docker container is running successfully (ID: 3c10bcf71a6d...).
Open Claude Desktop → Settings → Developer → Edit Config and add this:## 📝 Setup Steps
Here are test cases from basic to advanced:
List all my HITL escalationsExpected: Should show your existing HITL escalations or empty list
Create a HITL escalation with these details:
- Session ID: "test-session-123"
- Agent ID: "claude-test-agent"
- User prompt: "User wants to purchase premium subscription"
- Proposed action: Purchase of "Premium Plan" for $29.99
- Escalate to: "[email protected]"
- Priority: "normal"Expected: Should return escalation ID like "esc-xxxxx-xxxxx-xxxxx"
Get details for HITL escalation "esc-[ID-FROM-TEST-2]"Replace [ID-FROM-TEST-2] with the actual ID from Test 2
Approve HITL escalation "esc-[ID-FROM-TEST-2]" with comment "Approved for testing purposes"Create a HITL escalation for a $50 software license purchase, then modify it to $75 enterprise licenseBefore testing in Claude, verify your container:
# Check container is running
docker ps | grep hitl-mcp-server
# Check container logs
docker logs hitl-mcp-server
# Test the container directly
docker exec -i hitl-mcp-server node -e 'console.log("✅ Container is working!")'If working correctly, Claude should:
esc-xxxxx-xxxxx-xxxxxdocker logs hitl-mcp-serverdocker ps | grep hitl-mcp-serverStart with Test 1 (list HITLs) - it's the safest and will confirm the connection is working!
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.