Remote AI Workspace

Connect any MCP-compatible AI — ChatGPT, Claude Desktop, Cursor — to your CQ knowledge base via mcp.pilab.kr.

Why Remote AI Workspace

CQ installed locally gives you 275+ MCP tools and GPU access. The Remote AI Workspace gives you the knowledge layer — accessible from any AI tool, on any device, without a local install.

• Knowledge recorded in Claude Code is available in ChatGPT
• Experiment results from your GPU server sync to your desktop
• Any AI can read and write to your shared workspace

Connect in 2 Steps

Step 1: Log in at mcp.pilab.kr

Visit https://mcp.pilab.kr and log in with GitHub. This creates your OAuth token.

Step 2: Add the MCP server to your AI tool

{
  "mcpServers": {
    "cq-brain": {
      "url": "https://mcp.pilab.kr/mcp",
      "type": "streamable-http"
    }
  }
}

That's it. GitHub OAuth handles authentication — the URL is the same for everyone; access is controlled by your token.

AI Tool Setup

Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "cq-brain": {
      "url": "https://mcp.pilab.kr/mcp",
      "type": "streamable-http"
    }
  }
}

Restart Claude Desktop.

ChatGPT (Custom GPT / Connector)

In ChatGPT's MCP connector settings, add:

URL: https://mcp.pilab.kr/mcp
Auth: OAuth 2.1 (GitHub)

Cursor

Add to .cursor/mcp.json in your project or ~/.cursor/mcp.json globally:

{
  "mcpServers": {
    "cq-brain": {
      "url": "https://mcp.pilab.kr/mcp",
      "type": "streamable-http"
    }
  }
}

What the Remote AI Workspace Can Do

The Remote MCP server exposes a subset of CQ tools focused on knowledge:

Tool	Description
cq_knowledge_record	Save a discovery, experiment result, or decision. Multi-project users get a project selector.
cq_knowledge_search	Search across all your projects — personal + team + global knowledge combined.
cq_session_summary	Capture a session summary to your knowledge base.
cq_status	View project status, tasks, and progress.

3-Tier Knowledge Search

When you search, CQ combines results from three layers:

Layer	Scope	Example
Personal	Your knowledge from any AI tool	"JWT refresh token design" you discussed in ChatGPT
Project	All team members' tagged knowledge	A colleague's API rate limit decision in Claude Code
Global	Anonymized community knowledge	Common Go error handling patterns from all CQ users

You don't choose which layer to search — all three are automatically combined. Results show their source (personal, team, or global).

These tools work identically across Claude Code, ChatGPT, and Cursor — same data, same backend.

AI Self-Capture

The Remote AI Workspace is engineered so AI tools proactively save knowledge without you asking.

Tool descriptions are written to trigger automatic saves:

• When an AI discovers a bug root cause → cq_knowledge_record fires automatically
• When a solution to a non-obvious problem is found → saved immediately
• When a pattern emerges across multiple files → recorded with context

You don't need to prompt "save this" — the AI does it when it recognizes something worth keeping.

Session Summary

When a session ends, the AI can call cq_session_summary to capture:

• Key decisions made
• Preferences expressed
• Problems solved and how
• Files changed and why

This feeds directly into the Knowledge Loop — preferences extracted from summaries accumulate into hints and rules.

OAuth Flow

Your AI tool → mcp.pilab.kr/mcp
                     │
               OAuth 2.1 (GitHub)
                     │
               Validate token
                     │
               Route to your Supabase namespace
                     │
               Return knowledge results

The Cloudflare Worker (mcp.pilab.kr) acts as an OAuth 2.1 proxy. It validates your GitHub token, identifies your user namespace in Supabase, and proxies knowledge operations. No CQ binary required on the remote machine.

Cross-Platform Knowledge Sync

When you record knowledge in any tool, it's immediately available in all others:

Claude Code session:  discovers caching bug root cause
  → cq_knowledge_record("Redis TTL not set for anonymous sessions")

Same day, ChatGPT:    "why does the cache behave inconsistently?"
  → cq_knowledge_search("cache") returns the earlier discovery

No copy-paste. No re-explanation. Knowledge follows you.

Requirements

• CQ account (free — sign up at mcp.pilab.kr)
• GitHub account for OAuth
• Any MCP-compatible AI tool

No local CQ installation required for the Remote AI Workspace connection.

Relationship to Local Installation

The Remote AI Workspace and local CQ use the same Supabase backend. They are not separate systems:

Feature	Local CQ	Remote AI Workspace
Task orchestration	Yes	No
GPU job execution	Yes	No
File access	Yes	No
Knowledge read/write	Yes	Yes
Knowledge Loop	Yes	Yes (via summary)
Setup required	Install + build	OAuth login

Use local CQ for building and GPU experiments. Use the Remote AI Workspace for knowledge access from tools where you can't install software.

Next Steps

• Knowledge Loop — how knowledge accumulates into preferences and rules
• Tiers — Remote AI Workspace is available in Pro and Team tiers