Documentation
Protocol Effective: July 15, 2026
Welcome to the AgentHelm documentation. AgentHelm provides a centralized control plane and a shared **Project Brain** for your autonomous AI agents. Learn how to configure, secure, and execute agent governance.
Installation & Connection
1. Python SDK
Install the official AgentHelm SDK from PyPI:
pip install agenthelm-sdk
Initialize and connect your agent:
from agenthelm import Agent
agent = Agent(
key="YOUR_CONNECT_KEY",
name="Analytics Agent",
project="AgentHelm Platform"
)2. Model Context Protocol (MCP)
Plug AgentHelm into Cursor, Claude Code, or Claude Desktop by adding the MCP server configuration:
{
"mcpServers": {
"agenthelm": {
"command": "npx",
"args": ["-y", "agenthelm-mcp"],
"env": {
"AGENTHELM_CONNECT_KEY": "YOUR_CONNECT_KEY",
"AGENTHELM_PROJECT": "my-project-name"
}
}
}
}Exposes tools: get_context, propose_knowledge, and get_history.
Governance Framework
AgentHelm enforces safety bounds using a classification model:
- Read-Only: Actions that query state without modification are passed automatically.
- Side Effect: Actions that modify state safely are tracked and logged with rate-limiting.
- Irreversible: Sensitive operations (deleting data, executing shell code) hold in a pending state until approved.
Full Technical Reference
For complete API specifications, compilation lifecycle descriptions, and database schema diagrams, visit our primary open-source repository readme:
View GitHub ReadmeAgent Governance Protocol — Secure Access Guaranteed