Agent Integration Overview
Configure prAxIs OS for your specific AI agent and IDE. Choose based on your primary development environment.
Installation Command Pattern​
"Install prAxIs OS from github.com/honeyhiveai/praxis-os for <AGENT> (optional: in <IDE>)"
Examples:
for Cursor→ Cursor IDE (primary, native MCP)for Cline in VS Code→ Cline extension in VS Code (primary)for Cline in Cursor→ Cline as secondary agent (connects to Cursor's MCP server)for Claude Code→ Claude Code CLI (primary, terminal) - default when no IDE specifiedfor Claude Code in VS Code→ Claude Code VS Code extension (primary) - when IDE specifiedfor GitHub Copilot→ GitHub Copilot (primary, VS Code, JetBrains, Xcode, Eclipse)
Branching Logic for Claude Code:
"for Claude Code"(no IDE specified) → Route toclaude-code/terminal.md(CLI/terminal mode)"for Claude Code in VS Code"(IDE specified) → Route toclaude-code/vscode.md(VS Code extension)- If user is IN VS Code but says
"for Claude Code"→ Default to CLI, but consider asking if they want VS Code extension
Primary vs Secondary Agents​
Primary Agents (Control MCP Server)​
Primary agents own the MCP server lifecycle - they start, stop, and manage it:
| Agent | IDE | MCP Connection | Guide |
|---|---|---|---|
| Cursor | Cursor IDE | Native MCP via .cursor/mcp.json | cursor/ |
| Cline | VS Code | MCP via VS Code settings | cline/vscode.md |
| Claude Code | Terminal (CLI) | Direct MCP server invocation | claude-code/terminal.md |
| Claude Code | VS Code | MCP via VS Code integration | claude-code/vscode.md |
| GitHub Copilot | VS Code, JetBrains, Xcode, Eclipse | MCP via mcp.json | github-copilot/ |
When to use primary mode:
- This is your main AI coding assistant
- You want full control over MCP server
- Single agent workflow
Secondary Agents (Connect via HTTP)​
Secondary agents connect to an existing MCP server via HTTP (run by a primary agent):
| Agent | Primary Agent | Connection | Guide |
|---|---|---|---|
| Cline | Cursor | HTTP to Cursor's MCP server | cline/cursor.md |
| Claude Code | Cursor | HTTP to Cursor's MCP server | claude-code/cursor.md |
When to use secondary mode:
- Multi-agent workflow (Cursor + Cline, for example)
- Primary agent already managing MCP server
- Want specialized agent for specific tasks
Example multi-agent setup:
- Cursor (primary) - Main development, owns MCP server
- Cline (secondary) - Connects to Cursor's MCP via HTTP for specialized tasks
Quick Decision Tree​
Do you use Cursor as your main IDE?
├─ YES → Primary: cursor/
│ └─ Want to add Cline? → Secondary: cline/cursor.md
│ └─ Want to add Claude Code? → Secondary: claude-code/cursor.md
│
└─ NO → Do you use VS Code?
├─ YES → Primary: cline/vscode.md, claude-code/vscode.md, or github-copilot/
│ └─ For Claude Code: Use "for Claude Code in VS Code" → claude-code/vscode.md
│ └─ For GitHub Copilot: See [GitHub Copilot guide](./github-copilot/)
└─ NO → Primary: claude-code/terminal.md (CLI)
└─ For Claude Code: Use "for Claude Code" → claude-code/terminal.md
Claude Code Branching:
"for Claude Code"(no IDE) →claude-code/terminal.md(CLI/terminal)"for Claude Code in VS Code"→claude-code/vscode.md(VS Code extension)
LLM Model Support​
All integrations work with multiple LLM models. Recommended:
- Claude 3.5 Sonnet - Best MCP tool calling, recommended
- GPT-4 Turbo - Good alternative
- Claude 3 Opus - Supported
- GPT-4 - Supported
Note: Tool calling quality varies by model. Claude 3.5 Sonnet has best performance with MCP.
MCP Server HTTP Mode (For Secondary Agents)​
When running secondary agents, the primary agent must use --transport dual to enable HTTP endpoint:
In .cursor/mcp.json (or equivalent for primary agent), use:
{
"mcpServers": {
"praxis-os": {
"command": "${workspaceFolder}/.praxis-os/venv/bin/python",
"args": [
"-m",
"ouroboros",
"--transport",
"dual",
"--log-level",
"INFO"
],
"env": {
"PROJECT_ROOT": "${workspaceFolder}",
"PYTHONPATH": "${workspaceFolder}/.praxis-os",
"PYTHONUNBUFFERED": "1"
}
}
}
}
Important: Primary agents always use --transport dual (enables HTTP for secondary agents).
Secondary agents then connect to the HTTP endpoint via URL (port allocated dynamically, written to .praxis-os/.mcp_server_state.json).
See specific secondary agent guides for detailed configuration.
Common Questions​
Can I use multiple primary agents?​
No - only one agent should control the MCP server. Use primary + secondary pattern instead.
Can I switch between agents?​
Yes, but only one primary agent at a time. Close the primary agent before starting another, or use secondary mode.
Do I need different instruction files for different agents?​
Yes - each agent has its own instruction file format:
- Cursor: Uses
.cursorrules(Cursor-specific) - Cline: Uses
.clinerules(Cline-specific) - Claude Code: Uses
.claude/CLAUDE.md(Claude Code-specific) - GitHub Copilot: Uses
.github/copilot-instructions.md(GitHub Copilot-specific)
What if my agent isn't listed?​
If your agent supports MCP protocol:
- Follow the closest guide (probably Cline in VS Code or Cursor)
- Adapt MCP configuration for your agent's format
- Core concept: Point agent to
.praxis-os/venv/bin/python -m ouroboros
Can GitHub Copilot use prAxIs OS?​
Yes! GitHub Copilot supports MCP protocol. See the GitHub Copilot integration guide for detailed setup instructions.
Supported IDEs:
- VS Code (via GitHub Copilot extension)
- JetBrains IDEs (IntelliJ IDEA, PyCharm, WebStorm, etc.)
- Xcode (via GitHub Copilot for Xcode extension)
- Eclipse (version 2024-09+)
Configuration:
GitHub Copilot uses mcp.json configuration files. You can configure via:
- MCP Registry (browse and install servers) - Easiest method
- Manual configuration in
mcp.json- For advanced setups
For detailed setup instructions, see the GitHub Copilot integration guide.
Next Steps​
- Choose your agent - Primary or secondary mode
- Follow the guide - Complete configuration for your agent
- Run orientation - Start with 10 bootstrap queries
- Start building - prAxIs OS provides guidance at every step
Getting Help​
- Installation issues: Check troubleshooting in your specific agent guide
- MCP server issues: See MCP Server Update Guide
- Workflow questions: See Understanding Workflows