gRPC传输插件

MCP gRPC Transport (Python)

A pluggable gRPC-based transport implementation for the Model Context Protocol (MCP) Python SDK V2.

This library replaces the default JSON-RPC-over-streams transport with a high-performance, stateless gRPC implementation.

[!WARNING] This package is currently in active development and exposes an experimental API. The APIs and Protobuf schema are subject to change and breaking modifications may occur as the Model Context Protocol V2 specification and Python SDK mature.

Dependencies

This package requires the following core dependencies:

• mcp (Model Context Protocol Python SDK): Pinned to version 2.1.1.
• grpcio: Required for asynchronous gRPC client and server communications (grpcio>=1.83.1).
• mcp-grpc-transport-proto: Generated Protobuf message and service contracts.

Features

• Pydantic Integration: Automatically integrates with the Python SDK V2 Pydantic models, converting snake_case attributes to generated Protobuf messages.
• Dual Server Support: Works with both the high-level MCPServer (via decorators like @mcp.tool()) and the low-level Server API.
• Managed & Unmanaged Lifecycles: Allows embedding the gRPC server inside your existing gRPC application (unmanaged mode) or running it standalone (managed mode).

Quick Start

Setting up the gRPC transport involves defining your server using the SDK's high-level MCPServer and connecting to it via GRPCClientDispatcher plugged into the ClientSession.

• Server Setup: See examples/simple_tool/server.py for a complete standalone server example.
• Client Setup: See examples/simple_tool/client.py to see how the client session connects and calls tools.

No MCP initialize handshake. This transport intentionally skips the MCP initialize / capability-negotiation step. Do not call await session.initialize() on a ClientSession bound to GRPCClientDispatcher; invoke tool/resource methods directly. Notifications (session.send_notification) are not supported either.

Refer to the Examples section below on how to run them.

Examples

We provide runnable examples demonstrating different MCP capabilities:

• Simple Tool Example (examples/simple_tool/): Demonstrates how to register a basic calculator tool using the high-level MCPServer decorators and call it from a client session.
• Simple Resource Example (examples/simple_resource/): Demonstrates registering static and templated text resources on the server and listing/reading them from the client.

To run any example:

1. Start the server:
   复制

   uv run python examples/<example_name>/server.py
   

2. In a separate terminal, run the client:
   复制

   uv run python examples/<example_name>/client.py
   

Documentation

• For details on the client-side design, see docs/client.md.
• For details on the server-side design, see docs/server.md.
• For developer guidelines, see AGENTS.md.

Development

This repository uses uv for managing python environments, formatting, and running tests.

Tests use absl.testing.absltest (for sync) and unittest.IsolatedAsyncioTestCase (for async). No pytest.

Run Tests

uv run python -m unittest discover -s tests -v

Run Tests with Coverage Report

uv run coverage run --source=src/mcp_grpc_transport -m unittest discover -s tests
uv run coverage report -m