concierge

Overview

Concierge acts as a wrapper for existing MCP servers, providing a fabric for building more sophisticated AI applications. It solves the problem of overwhelming AI agents with too many tools at once by implementing progressive disclosure, where only relevant tools are exposed based on the current workflow step. The framework also enables state management between steps, tool ordering through defined transitions, and semantic search for APIs with hundreds of tools. It maintains full protocol compatibility while adding these powerful features.

Try asking AI

After installing, here are 5 things you can ask your AI assistant:

• search_productssearch_products(query: str) -> dict

  Search the product catalog.

  Try:Find laptops under $1000Try:Show me all available products
• add_to_cartadd_to_cart(product_id: str) -> dict

  Add a product to the cart.

  Try:Add product p1 to cartTry:Put the selected item in my cart
• checkoutcheckout(payment_method: str) -> dict

  Complete the purchase.

  Try:Complete purchase with credit cardTry:Finish the order process
• search_toolssearch_tools(query: str) -> dict

  Find tools by description.

  Try:Find tools related to paymentTry:Show me all inventory functions
• call_toolcall_tool(tool_name: str, args: dict) -> dict

  Execute a discovered tool.

  Try:Execute the view_product functionTry:Run the add_to_cart tool

Note: Inferred tools from example code in the README. The actual tools depend on what the developer implements with the Concierge SDK.

Comparable tools

Installation

pip install concierge-sdk

concierge init my-project
cd my-project
python main.py

from concierge import Concierge
from mcp.server.fastmcp import FastMCP
app = Concierge(FastMCP("my-server"))

{
  "mcpServers": {
    "concierge": {
      "command": "python",
      "args": ["-m", "concierge", "run"],
      "env": {}
    }
  }
}

FAQ

How does Concierge differ from regular MCP servers?

Concierge adds progressive tool disclosure, workflow state management, and semantic search capabilities on top of standard MCP servers. It acts as a wrapper that enhances existing servers without requiring code changes to your tool definitions.

Can I use Concierge with MCP servers written in other languages?

Yes, since Concierge works at the MCP protocol level, it can wrap any MCP server regardless of the implementation language, as long as it adheres to the MCP specification.

Compare concierge with