mcp-mesh 2.2.1

Python SDK for MCP Mesh — build distributed AI agents with auto-discovery, dependency injection, and LLM integration

MCP Mesh Python Runtime

Python runtime for the MCP Mesh service mesh framework.

Installation

pip install mcp-mesh

Quick Start

import mesh

# Import types from public API
from mesh.types import McpMeshTool

# Define your agent
@mesh.agent(name="hello-world", http_port=9090)
class HelloWorldAgent:
    """Hello World agent demonstrating MCP Mesh features."""
    pass

# Create a greeting function with dependency injection
@mesh.tool(
    capability="greeting",
    dependencies=["date_service"],
    description="Greeting function with date dependency injection"
)
def greet(name: str = "World", date_tool: McpMeshTool = None) -> str:
    """Greeting function with automatic dependency injection."""
    if date_tool is not None:
        try:
            current_date = date_tool()
            return f"Hello, {name}! Today is {current_date}"
        except Exception:
            pass

    return f"Hello, {name}!"

# The runtime auto-initializes when you import mcp_mesh
# Your functions are automatically registered with the mesh registry

Features

• Automatic Registration: Functions are automatically registered with the Go registry
• Health Monitoring: Built-in health checks and heartbeats
• Dependency Injection: Inject dependencies into your functions
• Service Discovery: Find and use other services in the mesh
• Graceful Degradation: Works even if registry is unavailable

Configuration

The runtime can be configured via environment variables:

• MCP_MESH_ENABLED: Enable/disable runtime (default: "true")
• MCP_MESH_REGISTRY_URL: Registry URL (default: "http://localhost:8080")
• MCP_MESH_AGENT_NAME: Custom agent name (auto-generated if not set)

API Architecture

MCP Mesh uses a clear separation between public and private APIs:

• mesh - Public user API for decorators and types
• _mcp_mesh - Private internal implementation (do not import directly)

The underscore prefix on _mcp_mesh follows Python conventions to indicate internal/private packages. Users should only import from the mesh package to ensure compatibility across versions.

Documentation

See the main repository for complete documentation.