C
Causalos MCP
作者 @CausalOS
MCP server by CausalOS
创建于 4/14/2026
更新于 about 4 hours ago
README
Repository documentation and setup instructions
causalos-mcp
Causal memory for coding agents. Prevents your agent from repeating the same catastrophic mistakes across sessions.
Install
-
Install the CausalOS Python library:
pip install causal-os -
Start the local causal server:
causal-os serve -
Add the MCP server to Claude Code:
claude mcp add causalos-mcp npx causalos-mcpOr add manually to
~/.claude/mcp.json:{ "mcpServers": { "causalos": { "command": "npx", "args": ["causalos-mcp"], "env": { "CAUSALOS_PORT": "7433", "CAUSALOS_AGENT_ID": "claude-code" } } } }
What it does
Before your agent deletes a database or runs a dangerous shell command, it checks if something similar caused damage before. After it acts, it records what happened. Next session, it knows.
Tools
causal_check— Risk score and historical context before any destructive actioncausal_record— Record outcome after an action completescausal_history— Audit trail of current or past sessionscausal_append_downstream— Add discovered consequences to past actionscausal_graph— Summary of the agent's full causal memory
Requirements
- Node.js 18+
- Python 3.10+
causal-osPython package installed and running (causal-os serve)
快速设置
此服务器的安装指南
安装包 (如果需要)
npx @modelcontextprotocol/server-causalos-mcp
Cursor 配置 (mcp.json)
{
"mcpServers": {
"causalos-causalos-mcp": {
"command": "npx",
"args": [
"causalos-causalos-mcp"
]
}
}
}