MCP Server Setup

What is MCP?

The Model Context Protocol (MCP) is an open standard developed by Anthropic for connecting AI assistants with external tools and services. It provides a standardized way for AI applications like Claude Desktop, Cursor, and other MCP-compatible clients to communicate with external servers that provide additional capabilities.

With MCP, you can extend Claude's capabilities by connecting it to databases, APIs, file systems, and specialized tools. The Plexor Labs MCP server exposes prompt optimization and multi-provider routing tools that help you save tokens and reduce costs while maintaining response quality.

How Plexor Labs MCP Server Works

The Plexor Labs MCP server runs as a subprocess spawned by Claude Desktop. It communicates via JSON-RPC over stdin/stdout, following the MCP specification. When you invoke a Plexor Labs tool in Claude Desktop, the request is sent to the local MCP server which processes it and returns the result.

The server provides five primary tools:

• Prompt Optimization - Compress prompts for reduced token usage while preserving meaning
• Prompt Analysis - Analyze prompts to understand optimization potential
• Statistics Tracking - Monitor your token savings across sessions
• Provider Routing - Route prompts to optimal CLI providers (Claude Code, Codex)
• Provider Listing - View available providers and their health metrics

Prerequisites

Before setting up the Plexor Labs MCP server, ensure you have:

• Python 3.11+ installed on your system
• uv package manager (recommended) or pip
• Claude Desktop application installed
• A Plexor Labs API key from your API Keys page

Installing uv (Recommended)

We recommend using uv for faster package installation. Install it with:

curl -LsSf https://astral.sh/uv/install.sh | sh

powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

Installation

Install the Plexor Labs package which includes the MCP server:

uv pip install plexor

pip install plexor

After installation, verify the MCP server is available:

which plexor-mcp  # Should output the path to plexor-mcp

Claude Desktop Configuration

To connect Plexor Labs to Claude Desktop, you need to add it to your MCP server configuration file.

Configuration Using uv (Recommended)

Using uv ensures the correct Python environment is used:

{
  "mcpServers": {
    "plexor": {
      "command": "uv",
      "args": [
        "run",
        "--with",
        "plexor",
        "plexor-mcp"
      ],
      "env": {
        "PLEXOR_API_KEY": "YOUR_API_KEY",
        "PLEXOR_BASE_URL": "https://api.plexor.dev"
      }
    }
  }
}

Configuration Using Direct Path

Alternatively, use the direct path to the plexor-mcp executable:

{
  "mcpServers": {
    "plexor": {
      "command": "/path/to/plexor-mcp",
      "args": [],
      "env": {
        "PLEXOR_API_KEY": "YOUR_API_KEY",
        "PLEXOR_BASE_URL": "https://api.plexor.dev"
      }
    }
  }
}

Environment Variables

The Plexor Labs MCP server supports the following environment variables for configuration:

Available MCP Tools

Once configured, the Plexor Labs MCP server exposes five tools to Claude Desktop:

Verification

After configuration, verify that the Plexor Labs MCP server is working correctly:

Example Usage in Claude Desktop

# You can ask Claude to use Plexor Labs tools:

User: Use the Plexor Labs optimize_prompt tool to optimize this prompt:
      "I would really appreciate it if you could please help me to write
       a Python function that calculates the factorial of a number"

Claude: I'll optimize that prompt using Plexor Labs.
        [Calls optimize_prompt tool]

        The optimized prompt is:
        "Write Python factorial function"

        Results:
        - Original tokens: 32
        - Optimized tokens: 5
        - Reduction: 84.4%
        - Tokens saved: 27

Troubleshooting

MCP Server Not Starting

Authentication Errors

Tools Not Appearing

Common Fixes

• JSON Syntax Errors: Use a JSON validator to check your config file
• Path Issues on Windows: Use forward slashes or escaped backslashes in paths
• Python Version: Ensure Python 3.11+ is installed and in PATH
• Package Not Installed: Run pip install --upgrade plexor

Debug Logging

Enable debug logging by setting the environment variable in your config:

{
  "mcpServers": {
    "plexor": {
      "command": "uv",
      "args": ["run", "--with", "plexor", "plexor-mcp"],
      "env": {
        "PLEXOR_API_KEY": "YOUR_API_KEY",
        "PLEXOR_LOG_LEVEL": "DEBUG"
      }
    }
  }
}

Advanced Configuration

Multiple MCP Servers

You can configure multiple MCP servers alongside Plexor Labs:

{
  "mcpServers": {
    "plexor": {
      "command": "uv",
      "args": ["run", "--with", "plexor", "plexor-mcp"],
      "env": {
        "PLEXOR_API_KEY": "YOUR_API_KEY"
      }
    },
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/files"]
    },
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_PERSONAL_ACCESS_TOKEN": "ghp_..."
      }
    }
  }
}

Self-Hosted Plexor Labs

If you're running a self-hosted Plexor Labs instance, update the base URL:

{
  "mcpServers": {
    "plexor": {
      "command": "plexor-mcp",
      "args": [],
      "env": {
        "PLEXOR_API_KEY": "YOUR_API_KEY",
        "PLEXOR_BASE_URL": "https://plexor.your-company.com"
      }
    }
  }
}

Default Routing Mode

Set the default routing mode for all optimization and routing operations:

Alternative: HTTP API Integration

If MCP doesn't fit your use case, you can integrate Plexor Labs directly via the HTTP API. This is useful for:

• Server-side applications
• Custom integrations not using Claude Desktop
• Direct API access without MCP overhead

See the HTTP API documentation for complete details.