home / mcp / nexus mcp server

Nexus MCP Server

Unifies multiple MCP servers and LLM providers behind a single endpoint for search and tool execution.

Installation

Add the following to your MCP client configuration file.

Configuration

{
  "mcpServers": {
    "grafbase-nexus": {
      "url": "https://api.githubcopilot.com/mcp/",
      "headers": {
        "PYTHONPATH": "/opt/mcp",
        "GITHUB_TOKEN": "YOUR_GITHUB_TOKEN"
      }
    }
  }
}

You set up a unified MCP gateway that aggregates multiple MCP servers, APIs, and LLM providers behind a single endpoint. It lets you route requests, search for tools across connected servers, and execute tools in a consistent and controlled way while enforcing security, rate limits, and observability.

How to use

You connect HTTP-based MCP servers and local STDIO MCP servers to Nexus and route all tool calls through a single interface. Use the search capability to discover tools across all connected servers and then execute the chosen tool through the same unified endpoint. The system handles authentication, tool discovery, and execution so you can focus on building your AI workflows and automations.

How to install

Prerequisites: ensure you have a supported runtime installed on your system. You will typically need a modern operating system with network access and a terminal.

# Quick setup: install the MCP gateway (example uses a package manager or prebuilt binary in your environment).
# Follow the exact steps shown in your provider’s guidance for your platform.

Configuration and running

Configure the MCP gateway to connect to your MCP servers. You can expose a HTTP-based remote MCP server as well as local STDIO servers. The following configuration examples show how to wire three servers: one HTTP MCP endpoint and two STDIO local servers.

# HTTP MCP server (remote)
[mcp.servers.github]
url = "https://api.githubcopilot.com/mcp/"
auth.token = "{{ env.GITHUB_TOKEN }}"

# STDIO MCP server (local) - filesystem based
[mcp.servers.filesystem]
cmd = ["npx", "-y", "@modelcontextprotocol/server-filesystem", "/home/YOUR_USERNAME/Desktop"]

# STDIO MCP server (local) - Python MCP server
[mcp.servers.python_server]
cmd = ["python", "-m", "mcp_server"]
env = { PYTHONPATH = "/opt/mcp" }
cwd = "/workspace"

Security and access control

Enable authentication and TLS, enforce CSRF protection, and implement rate limiting to protect your MCP gateway and downstream servers. Use OAuth2 where appropriate and restrict access with access control rules per server and per tool.

Notes and examples

All server connections are surfaced under the MCP endpoint. Tools discovered from downstream servers are namespaced with their server name, so you can call them in a provider/model format like github__search_code or filesystem__read_file.

Troubleshooting and tips

If a server fails to start, check the command and working directory specified in the configuration. Empty tool lists mean the server has not initialized tools yet. Review logs for errors and ensure environment variables are set as expected.

Additional configuration options

The following options show how to configure rate limits, token forwarding, TLS, and headers for MCP servers. Use the samples to tailor security and performance to your environment.

Tools and endpoints

Nexus provides two primary MCP tools you use from clients: search to discover tools contextually across connected servers and execute to run a selected tool with specified parameters. You can invoke tools from any connected MCP server and the results are routed back through the unified endpoint.

Available tools

search

Context-aware tool search across all connected MCP servers using natural language queries.

execute

Execute a specific tool with the provided parameters across any connected MCP server.

get_weather

Fetch current weather for a location using a tool defined in an MCP server example.