Avaly-ai-Corp/Avaly-MCP-server

A collection of containerised services that deliver AI-security tooling, a chat-based agent layer, and a Model Context Protocol (MCP). This collection contains an agentic version of TextAttack and Garak which can be executed via the bundled agent. Instructions and examples follow.

Features:

• Model Context Protocol Compliant With The MCP Spec.
• SSE Transport Mechanism
• Agentic Versions Of TextAttack And Garak
• MCP Server Is Fully Extensibe To Add More Tools.

Prerequisites

• Docker 20.10+ and the Docker Compose v2 CLI plugin.
• GPU (optional) if you want CUDA-accelerated TextAttack/Garak runs.
• Ollama running locally for LLM back-end.

Pull the base LLM once:

ollama pull magistral:24b-small-2506-q8_0

Quick Start

# Build & start every service (three terminals or run in detached mode)
docker compose -f mcp-docker-compose.yml   up --build -d
docker compose -f toolx-docker-compose.yml up --build -d
docker compose -f agent-docker-compose.yml up --build -d

The stack will be reachable at:

• http://localhost:7070/sse – MCP over Server-Sent Events
• http://localhost:7575 – ToolX micro-service
• http://localhost:5000 – Agent REST API

Endpoints

Agent API (service-agent)

Usage Examples

Note: Persist same 'userId' across all requests

# 1. Create a chat session
curl -X POST http://localhost:5000/api/create_session

# 2. Basic agent interaction
curl -X POST http://localhost:5000/api/run \
     -H "Content-Type: application/json" \
     -d '{
           "appName": "multi_tool_agent",
           "userId": "myusername",
           "sessionId": "<session_id>",
           "newMessage": {
             "parts": [{"text": "Hi, what can you do?"}],
             "role": "user"
           },
           "streaming": false
         }'

# 3. Launch a TextAttack evaluation
curl -X POST http://localhost:5000/api/run \
     -H "Content-Type: application/json" \
     -d '{
           "appName": "multi_tool_agent",
           "userId": "myusername",
           "sessionId": "<session_id>",
           "newMessage": {
             "parts": [{"text": "Please run an attack with model distilbert-base-uncased-finetuned-sst-2-english on the setfit/sst2 dataset using the textfooler recipe for 5 examples"}],
             "role": "user"
           },
           "streaming": false
         }'

# 4. Launch a Garak probe
curl -X POST http://localhost:5000/api/run \
     -H "Content-Type: application/json" \
     -d '{
           "appName": "multi_tool_agent",
           "userId": "myusername",
           "sessionId": "<session_id>",
           "newMessage": {
             "parts": [{"text": "Please run a scan with model gpt2 and probe xss"}],
             "role": "user"
           },
           "streaming": false
         }'

Sample Output

Architecture

All three services are defined in their own Docker Compose files to allow independent builds. For production deployments you can spin them up concurrently or cherry-pick only the components you need.

┌──────────────┐   SSE         ┌───────────────┐   REST   ┌──────────────┐
│  Clients     │──────────────►│ service-mcp   │─────────►│service-toolx │
└──────────────┘               └───────────────┘          └──────────────┘
                                    │
                                    │ REST
                                    ▼
                               ┌───────────────┐
                               │ service-agent │
                               └───────────────┘

Contributions

Contributions welcome! 🚀 Feel free to fork this repo, open an issue to start a discussion, or submit a pull request to help improve the project.

License

This repository is licensed under the Apache 2.0 license. The license is avaialable at

Credits

Built with ❤️ by Avaly.ai. For support, reach out at: contact@avaly.ai.