Universal AI Chat

You set up a cross-platform MCP server that enables real-time, multi-session chat among Claude Code, OpenAI Codex CLI, and Google Gemini CLI. It provides shared memory for a common vector space, conversation history, and broadcast or collaboration requests across AI platforms, so you can orchestrate complex AI workflows in a single, coherent system.

How to use

Connect your three AI clients to the Universal AI Chat MCP Server to enable real-time, cross-platform conversations. Each platform registers as its own session and shares a common vector memory, so messages, context, and collaboration requests propagate across Claude Code, Codex CLI, and Gemini CLI. You can register multiple Claude Code sessions, broadcast messages to all connected AIs, and store/ retrieve shared context for coordinated tasks.

Common workflows include registering sessions, sending messages between platforms, checking for new messages, and retrieving full conversation history. You can also store shared context that all AIs can inspect and update, enabling coordinated goal-driven interactions across tools.

How to install

# Prerequisites
- Python 3.8+
- pip

# Install in development mode with vector support (optional)
pip install -e .
pip install -e ".[vector]"

# Run tests
pytest

Next, configure each MCP client to connect to the server using the provided command and environment variables. The following configurations are shown as examples for each platform.

# Claude Code
# Add to ~/.claude.json
"universal-ai-chat": {
  "command": "python3",
  "args": ["-m", "universal_ai_chat.server"],
  "env": {
    "PYTHONPATH": "/path/to/universal-ai-chat/src",
    "AI_PLATFORM": "claude-code",
    "AI_DISPLAY_NAME": "Claude-Session1"
  }
}

# OpenAI Codex CLI
# Add to ~/.codex/config.toml
[mcp_servers.universal-ai-chat]
command = "python3"
args = ["-m", "universal_ai_chat.server"]

[mcp_servers.universal-ai-chat.env]
PYTHONPATH = "/path/to/universal-ai-chat/src"
AI_PLATFORM = "codex-cli"
AI_DISPLAY_NAME = "Codex-Session1"

# Google Gemini CLI
# Add to ~/.gemini/settings.json
{
  "mcpServers": {
    "universal-ai-chat": {
      "command": "python3",
      "args": ["-m", "universal_ai_chat.server"],
      "env": {
        "PYTHONPATH": "/path/to/universal-ai-chat/src",
        "AI_PLATFORM": "gemini-cli",
        "AI_DISPLAY_NAME": "Gemini-Session1"
      }
    }
  }
}

Additional notes

Environment variables define platform types, session names, and how the server locates its libraries and memory. The central settings include the Qdrant vector memory, the base storage path, and the Qdrant host/port. You can adjust these values to fit your deployment environment.

Key environment variables include AI_PLATFORM to identify the source platform (claude-code, codex-cli, gemini-cli), AI_DISPLAY_NAME for readable session names, AI_SESSION_ID for unique IDs, NODE_ID for cluster identification, STORAGE_BASE as the base path for databases, and QDRANT_HOST/PORT for the vector store.

Usage overview and tools

You can register sessions, list active sessions, send and broadcast messages, check for new messages, and retrieve conversation history. You can also set and get shared context that all connected AIs can access, and you can issue collaboration requests to specific AI platforms.

Available tools

register_session

Register a new AI session with the MCP system.

list_active_sessions

List all currently connected Claude Code, Codex CLI, and Gemini sessions.

send_message

Send a message from one session to another specific session.

broadcast_message

Broadcast a message to all connected sessions.

check_messages

Check for new messages in your session inbox.

get_conversation

Retrieve the full conversation history for a session.

set_shared_context

Store a shared context value accessible by all sessions.

get_shared_context

Retrieve a shared context value.

request_collaboration

Request assistance from a specific AI platform.

get_platform_info

Show supported AI platforms.