strata-mcp 1.0.0

One Model Context Protocol (MCP) server for AI agents to handle thousands of tools

Overview

Strata is a unified MCP server that intelligently manages thousands of tools across multiple applications, presenting them to AI agents progressively without overwhelming the model context.

🚀 Looking for production-ready solution? We offer a fully managed Strata service with enterprise-grade performance, native OpenAPI parsing, and advanced AI model optimizations. Get started instantly →

This repository contains our open-source implementation - a streamlined version you can easily deploy as your MCP aggregation manager in your local infrastructure.

🚀 The Problem We Solve

Imagine you have:

• 📦 Dozens of MCP servers, each with dozens of tools
• 🤖 AI models seeing 100s-1000s of tools at once
• 📉 Degraded performance due to massive context

✨ The Strata Solution

Instead of flooding your AI model with hundreds of tools, Strata acts as an intelligent router:

🔍 Smart Discovery → Only exposes a few essential tools to the model
🎯 Progressive Access → Finds and surfaces the right tool when needed
⚡ Optimized Context → Maintains peak model performance
🔗 Seamless Integration → Works with your existing MCP servers, and easy configuration

Your AI agent gets access to thousands of tools while maintaining high performance! ⚡

Quick Start

Installation

pipx install strata-mcp

Or with pip:

pip install strata-mcp

For development:

pip install -e .

Configure MCP Servers

You can configure your MCP servers by strata CLI tool. Or manually configure it in a JSON file, just like your other MCP JSON config files.

Add MCP servers

• Stdio Server

strata add --type stdio <server_name> npx @playwright/mcp@latest

• SSE Server

strata add --type sse <server_name> http://localhost:8080/mcp/ --env API_KEY=your_key

• HTTP Server

strata add --type http <server_name> https://api.githubcopilot.com/mcp/ --header "Authorization=Bearer token"

• HTTP Server with OAuth

strata add --type http <server_name> https://mcp.notion.com/mcp --auth_type oauth

List Servers

strata list

Enable/Disable Servers

strata enable <server_name>
strata disable <server_name>

Remove Servers

strata remove server-name

Manual Configuration

Configuration is stored in ~/.config/strata/servers.json by default. You can specify a custom config path:

strata --config-path /path/to/config.json add --type stdio ...

Config Format

{
  "mcp": {
    "servers": {
      "github": {
        "command": "npx",
        "args": ["-y", "@modelcontextprotocol/server-github"],
        "env": {
          "GITHUB_TOKEN": "your_token"
        },
        "enabled": true
      },
      "api-server": {
        "type": "http",
        "url": "https://api.example.com/mcp",
        "headers": {
          "Authorization": "Bearer token"
        },
        "enabled": true
      }
    }
  }
}

Environment Variables

• MCP_CONFIG_PATH - Custom config file path
• MCP_ROUTER_PORT - Default port for HTTP/SSE server (default: 8080)

Running Strata MCP servers

Strata itself is a MCP server. You can run and use it like a normal MCP, in stdio mode or http/sse mode.

Stdio Mode (Default)

Run without arguments to start in stdio mode for direct MCP communication:

python -m strata
# or
strata

HTTP/SSE Server Mode

Run with port to start as HTTP/SSE server:

strata run --port 8080

Tool Integration

Strata can automatically configure itself in various AI assistants and IDEs that support MCP.

Add Strata to Claude Code

# Add to user configuration (default)
strata tool add claude

# Add to project-specific configuration
strata tool add claude --scope project

Add Strata to Gemini

strata tool add gemini

Add Strata to VSCode

strata tool add vscode

Add Strata to Cursor

# Add to user configuration (~/.cursor/mcp.json)
strata tool add cursor --scope user

# Add to project configuration (.cursor/mcp.json)
strata tool add cursor --scope project

Supported scopes:

• user: Global configuration (default)
• project: Project-specific configuration
• local: Same as project (for Cursor)

Note: VSCode doesn't support scope parameter and will use its default behavior.

Available Tools

When running as a router, the following tools are exposed:

• discover_server_actions - Discover available actions from configured servers
• get_action_details - Get detailed information about a specific action
• execute_action - Execute an action on a target server
• search_documentation - Search server documentation
• handle_auth_failure - Handle authentication issues

Development

Running Tests

pytest

Project Structure

• src/strata/ - Main source code
  • cli.py - Command-line interface
  • server.py - Server implementation (stdio/HTTP/SSE)
  • tools.py - Tool implementations
  • mcp_client_manager.py - MCP client management
  • config.py - Configuration management

Examples

Running GitHub MCP Server through Router

# Add GitHub server (official HTTP server)
strata add --type http github https://api.githubcopilot.com/mcp/

# Run router in stdio mode
strata

# Or run as HTTP server
strata run --port 8080

Running Multiple Servers

# Add multiple servers
strata add --type stdio playwright npx @playwright/mcp@latest
strata add --type http github https://api.githubcopilot.com/mcp/

# List all servers
strata list

# Run router with all enabled servers
strata run --port 8080

🚀 Quicker Start

Skip the configuration hassle! 🎯

Don't want to spend time configuring dozens of MCP servers? We've got you covered!

✨ What You Get Instantly:

• 🏗️ Fully Hosted Strata - Zero setup, production-ready
• 🛠️ 40+ Pre-configured MCP Servers - Ready to use out of the box
• 🌐 Multi-Domain Coverage - Development, productivity, APIs, databases, and more
• ⚡ Thousands of Tools - At your fingertips immediately

💡 Perfect For:

• 🚀 Rapid Prototyping - Start building with AI tools in minutes
• 🧪 Testing & Exploration - Try Strata without any commitment
• 💼 Production Workloads - Enterprise-grade reliability

🎁 Get Started in 1 minute:

Just point your AI assistant to our hosted Strata endpoint and you're ready to go!

---

Ready to experience the power of thousands of AI tools?

👉 Get Instant Access at Klavis AI (YC X25) 👈

From zero to thousands of tools in under a minute! ⏱️

License

Apache License 2.0