nexus -extension
Nexus-MCP solves Tool Space Interference in LLMs. It uses a single gateway to aggregate multiple MCP servers and employs a deterministic 4-phase workflow (Discovery, Mapping, Schema Verification, Bridged Execution) to prevent context saturation and tool hallucinations, maintaining reasoning accuracy with massive tool ecosystems.
claude mcp add --transport stdio tanaikech-nexus-mcp-extension node /path/to/nexus-mcp-extension/mcp-server/src/nexus-mcp.js \ --env MCP_SERVER_LIST="Path to mcp_config.json (e.g., /home/user/mcp_config.json)"
How to use
Nexus-MCP acts as a single gateway that aggregates multiple MCP servers deployed as Gemini extensions. It enforces a deterministic 4-phase workflow: Discovery (collect all available tools from connected servers), Mapping (identify the target SERVER_ID and TOOL_ID), Schema Verification (fetch the technical schema only for the chosen tool), and Bridged Execution (route the command to the appropriate downstream server). Through Nexus-MCP you can introspect the available tools, route specific tasks to preferred servers, and chain tool usage across multiple servers to accomplish complex workflows such as document generation, conversion, and dissemination. You interact with the gateway via Gemini CLI or Google Antigravity, which both leverage the Nexus gateway under the hood. The gateway does not implement tools itself; it manages downstream MCP servers and coordinates tool execution across them, reducing tool-space interference and maintaining stable context flow for LLM agents.
How to install
Prerequisites:
- Node.js (Latest LTS) or the environment where your Gemini CLI extension runs
- Gemini CLI installed and authenticated
- (Optional) Google Antigravity for enhanced agentic workflows
- Install the Nexus-MCP extension for Gemini CLI:
gemini extensions install https://github.com/tanaikech/nexus-mcp-extension
- Prepare a configuration file named mcp_config.json (example provided in the repository). This file should list downstream MCP servers with their commands and arguments. Example:
{
"mcpServers": {
"google-workspace": {
"command": "node",
"args": [
"/home/username/.gemini/extensions/google-workspace/dist/index.js"
]
},
"tools-for-mcp-server": {
"command": "node",
"args": [
"/home/username/.gemini/extensions/tools-for-mcp-server-extension/mcp-server/src/index.js"
],
"timeout": 300000
}
}
}
- Point Nexus-MCP to the configuration file by exporting the MCP_SERVER_LIST environment variable:
export MCP_SERVER_LIST="/home/username/mcp_config.json"
On Windows PowerShell:
$env:MCP_SERVER_LIST="C:\Users\username\mcp_config.json"
- Disable downstream extensions in Gemini CLI so they are accessed only through Nexus-MCP:
gemini extensions disable google-workspace
gemini extensions disable tools-for-mcp-server
- (Optional) If using Antigravity, ensure the Nexus-MCP server is registered as described in the Antigravity guidance and that the MCP_SERVER_LIST points to your config.
Additional notes
Tips and common issues:
- Ensure the MCP_SERVER_LIST points to a valid JSON file and that each downstream server uses absolute paths to its executable script.
- If a downstream tool times out, you can increase the timeout in the server entry (e.g., "timeout": 600000).
- When routing, you can explicitly specify the target server to the LLM agent (e.g., by including a route hint like Use google-workspace) to optimize tool selection.
- The gateway relies on the underlying Gemini extension environment; keep extensions updated and ensure required authentication (e.g., Gemini, Google Antigravity) is valid.
- The 4-phase workflow helps maintain deterministic reasoning; ensure your agents follow the routing order to avoid tool-space interference.
Related MCP Servers
mcp-for-beginners
This open-source curriculum introduces the fundamentals of Model Context Protocol (MCP) through real-world, cross-language examples in .NET, Java, TypeScript, JavaScript, Rust and Python. Designed for developers, it focuses on practical techniques for building modular, scalable, and secure AI workflows from session setup to service orchestration.
desktop
Agentify Desktop lets Codex/Claude/OpenCode control your logged-in ChatGPT, Claude, AiStudio, Gemini, Grok, Perplexity web sessions via MCP, parallel hidden/visible tabs, file upload + image download
mcp-arr
MCP server for *arr media management suite
ToolsForMCPServer-extension
Simplified Google Workspace Automation with Gemini CLI Extensions
stitch -auto
Automated installer for Stitch MCP - The easiest way to set up your Universal MCP server for Google Stitch.
mcp-json-yaml-toml
A structured data reader and writer like 'jq' and 'yq' for AI Agents