MCP Server Setup

Paramus exposes computational chemistry tools to AI assistants through the Model Context Protocol (MCP). The tray application runs locally — no cloud services, no API keys, no manual configuration files.


Quick Start

Make sure the Paramus tray application is running (system tray icon visible).

VS Code (Copilot)

Add .vscode/mcp.json to your workspace:

{
  "servers": {
    "paramus-chemistry": {
      "type": "http",
      "url": "http://localhost:8765/mcp"
    }
  }
}

VS Code detects the server and asks to start it:

VS Code detects the MCP server and asks to start it

On first run, select Always Allow Tools from paramus-chemistry-os:

Trust dialog with permission options

Tools are now available. Ask any question and the AI calls them automatically:

Asking

Claude Desktop

Edit %APPDATA%\Claude\claude_desktop_config.json:

{
  "mcpServers": {
    "Paramus Chemistry": {
      "command": "C:\\Users\\<user>\\AppData\\Local\\Paramus\\Server\\paramus-client\\paramus-client.exe"
    }
  }
}

Restart Claude Desktop after saving.

Claude Desktop configuration and tool listing
Claude executing a Paramus tool call

Windsurf

Edit ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "Paramus Chemistry": {
      "command": "C:\\Users\\<user>\\AppData\\Local\\Paramus\\Server\\paramus-client\\paramus-client.exe"
    }
  }
}

Troubleshooting

ProblemSolution
VS Code shows no toolsCheck .vscode/mcp.json points to http://localhost:8765/mcp and tray icon is visible
Claude Desktop shows no toolsCheck claude_desktop_config.json path. Restart Claude after config changes
“Tray App not reachable”Start the tray application (paramus-app.exe or run.bat)
Tools missing after updateRestart the tray app (it auto-discovers tools on startup)
BRAIN/WORLD tools not showingCheck network connectivity to backend hosts. Unreachable services are hidden automatically

Details

Architecture

The tray application starts an HTTP server on localhost:8765. AI clients connect to this endpoint and receive access to 10 gateway meta-tools that route to the full tool catalog across 15 categories.

MCP Server Architecture

Tool Categories

Molecular Chemistry, Analytical Chemistry, Chemical Kinetics, Thermodynamics, Quantum Chemistry, Polymer Science, Materials Science, Electrochemistry, Drug Discovery, Data Science, Engineering, BRAIN Platform, WORLD Platform, OPERATE Platform, INTENT Core.

The visible count depends on which backend services (BRAIN, WORLD, OPERATE) are reachable.

Transport Modes

ClientTransportSubprocess?
VS CodeHTTP direct to :8765No
Claude DesktopSTDIO gateway -> HTTP proxy to :8765Yes
WindsurfSTDIO gateway -> HTTP proxy to :8765Yes

VS Code connects directly via HTTP. Claude Desktop and Windsurf use STDIO, so the paramus-client.exe (1.7 MB, stdlib-only, < 0.3s startup) bridges STDIO to HTTP. When running from source, use python -m paramus.intent.gateway instead.

Island Mode

When BRAIN or WORLD backends are unreachable, their tools are automatically hidden. They reappear when connectivity is restored. Local tools (RDKit, CoolProp, Cantera) always work.

Ports

PortService
8765Tray app MCP endpoint
8766Settings UI (browser)

Frequently Asked Questions

What is the Paramus MCP Server?

A local server implementing the Model Context Protocol standard. It exposes computational chemistry tools to AI assistants through 10 gateway meta-tools that route to the full catalog.

Which AI clients are supported?

VS Code (Copilot), Claude Desktop, Windsurf, ChatGPT Desktop, and any MCP-compatible client. VS Code connects via HTTP. Claude Desktop and Windsurf use STDIO via a thin gateway binary.

How do I install it?

Install Paramus Intent MCP. Start the tray application (or enable auto-start during installation). Add the MCP configuration to your AI client (see Quick Start). No API keys required for core tools. Optional keys can be configured in Settings for Materials Project and Rowan services.

Does it work offline?

Local tools (chemistry, polymers, thermodynamics, data science) work fully offline. BRAIN, WORLD, and OPERATE tools require network access to their respective backend servers.

Is it open source?

The server is distributed under a proprietary license (Paramus Transform LLC). It implements the open MCP protocol standard, allowing any MCP-compatible AI client to connect.

Scroll to Top