brendon92/agent-mcp-server
If you are the rightful owner of agent-mcp-server and would like to certify it and/or have it hosted online, please leave a comment on the right or send an email to dayong@mcphub.com.
The MCP-Server project implements a Multi-Tool Integration Architecture using the Model Context Protocol (MCP), providing a robust framework for managing and executing various tools through a web-based interface.
Agent MCP Server
A production-ready Model Context Protocol server designed for complex, agentic AI workflows.
Features
- Secure by Default: Enforced environment-based auth, input sanitization, and sandboxed code execution.
- Agentic Capabilities: Supports "Reasoning" features like Human-in-the-Loop (
ask_human) and reactive filesystem notifications. - Observability: Structured JSON logging with trace IDs and rate limiting.
- Check
logs/backend_server.log: You should see JSON log lines. - Modular Architecture: Pluggable Executors (Docker/Local) and clear tool separation.
Usage
Prerequisites
- Docker (recommended for sandboxing)
- Python 3.11+
- An MCP Client (e.g.,
forgery-ai-agentor Claude Desktop)
Quick Start (Docker)
-
Create a
.envfile:MCP_AUTH_TOKEN=your-secure-token-here -
Run with Docker Compose:
docker-compose up backend -
Connect your agent to the stdio or SSE endpoint (depending on deployment).
Configuration
Configuration is managed via Environment Variables (Pydantic validated).
| Variable | Description | Default |
|---|---|---|
MCP_AUTH_TOKEN | Required. Secret token for auth. | - |
MCP_SANDBOX_ENABLED | Enable Docker-based isolation. | True |
MCP_WORKSPACE_DIR | Root directory for file operations. | ./test_workspace |
Architecture
See for detailed diagrams and decision logs.
Security
- Sandboxing: Code execution runs in isolated Docker containers by default.
- Path Traversal Prevention: All file operations are jailed to the workspace.
Development
Run conformance tests:
pytest tests/conformance
See for roadmap.