pydantic-ai-resilient-mcp

Crash-resilient wrapper for Pydantic AI MCP servers. Handles server crashes gracefully without breaking your agent.

Installation

pip install pydantic-ai-resilient-mcp

Quick Start

from pydantic_ai import Agent
from pydantic_ai_resilient_mcp import ResilientMCPSSE

# Wrap your SSE MCP server
server = ResilientMCPSSE('http://localhost:8000/sse')

async with server:
    agent = Agent(model, toolsets=[server])

    # Use normally - crashes are handled automatically
    result = await agent.run('Your prompt here')
    print(result.output)

What It Does

When your MCP server crashes:

• ✅ Catches the crash gracefully
• ✅ Returns clean error message: {"error": "Host server crashed. Cannot use tools anymore"}
• ✅ Prevents infinite retry loops
• ✅ Agent continues running instead of crashing

Example

from pydantic_ai import Agent
from pydantic_ai.models.openai import OpenAIChatModel
from pydantic_ai.providers.openai import OpenAIProvider
from resilient_mcp import ResilientMCPSSE

model = OpenAIChatModel('gpt-4o', provider=OpenAIProvider(api_key='...'))
server = ResilientMCPSSE('http://your-server:8000/sse')

async with server:
    agent = Agent(model, toolsets=[server])

    # Works before crash
    result1 = await agent.run('Use a tool')
    print(result1.output)

    # Server crashes during this call - handled gracefully
    result2 = await agent.run('Use another tool')
    print(result2.output)  # {"error": "Host server crashed..."}

    # No infinite retries - agent continues cleanly

License

MIT