<2-3 sentences summarising what the project addresses and solves, the target audience, and the benefit.> Concierge addresses the complexity of exposing AI agent tools by implementing the Model Context Protocol (MCP). It provides a reliability layer for MCP servers, enabling progressive disclosure of relevant tools rather than a flat list. This approach benefits developers by guaranteeing deterministic results, reliable tool invocation, reduced LLM context window usage, and ultimately lower operational costs for AI applications.

How It Works

Concierge acts as a fabric for MCP servers, dynamically altering the tools/list response based on the current workflow step. This progressive disclosure ensures agents only see pertinent tools, simplifying interaction and reducing prompt costs. Key mechanisms include defining stages to group related tools and transitions to enforce workflow logic. It also supports server-side shared state management across steps and offers semantic search to collapse large toolsets into discoverable meta-tools, abstracting away API complexity.

Quick Start & Requirements

• Installation: pip install concierge-sdk (Python 3.9+ required; uv recommended for speed).
• Scaffolding: Use concierge init <project-name> to generate a starter project.
• Running: Execute python main.py for scaffolded projects or wrap existing MCP servers with from concierge import Concierge; app = Concierge(FastMCP("my-server")).
• Documentation: Full guides available at https://docs.getconcierge.app.

Highlighted Details

• Progressive Disclosure: Dynamically reveals only necessary tools per workflow step, minimizing agent confusion and LLM costs.
• Enforced Tool Ordering: Manages workflow progression via defined stages and transitions, ensuring adherence to business logic.
• Shared State: Enables server-side state persistence across workflow steps, eliminating LLM round-trips for data transfer.
• Semantic Search: Collapses extensive tool inventories into two meta-tools (search_tools, call_tool) for agent discoverability.
• Protocol Compatible: Seamlessly wraps any existing MCP server with minimal code modifications.
• Multiple Transports: Supports stdio, streamable HTTP, and SSE for flexible deployment across various environments.
• Scaffolding CLI: concierge init command generates a functional project structure, accelerating initial setup.

Maintenance & Community

• Community: Active Discord server available for support and discussion (https://discord.gg/bfT3VkhF).
• Documentation: Comprehensive resources provided at https://docs.getconcierge.app.

Licensing & Compatibility

• The project's license is not explicitly stated in the README, which requires further investigation for commercial or derivative use.

Limitations & Caveats

• The README does not detail specific limitations, alpha/beta status, or known bugs.
• Functionality is predicated on the Model Context Protocol (MCP).
• Adoption involves understanding and potentially configuring stages, transitions, and state management for full workflow control.