# Integrations
QMCP integrates with external libraries to provide a complete agent development stack while avoiding duplication of maintained functionality.
## Philosophy
QMCP follows a "best of both worlds" approach:
1. **Use PydanticAI** for agent runtime, typed tools, and output validation
2. **Keep QMCP** for audit trails, HITL, model metadata, and multi-agent topologies
This hybrid architecture leverages actively-maintained libraries while preserving QMCP's unique production features.
## Available Integrations
### [PydanticAI](pydantic-ai.md)
LLM agent framework for Python. Provides:
- Agent execution with dependency injection
- Typed tool definitions with auto-generated schemas
- Output validation against Pydantic models
- Streaming and retry support
- Message history management
**Status**: Production-ready
```python
from qmcp.integrations.pydantic_ai import create_agent, QMCPToolset
agent = create_agent(
Models.CLAUDE_SONNET_4,
toolsets=[QMCPToolset()],
)
```
## Decision Matrix
When to use what:
| Need | Use |
|------|-----|
| Run an LLM agent | PydanticAI via `create_agent()` |
| Define typed tools for agents | PydanticAI `@agent.tool` |
| Expose tools via HTTP | QMCP `tool_registry` |
| Audit all tool calls | QMCP MCP Server |
| Human approval workflow | QMCP HITL API |
| Model pricing/limits | QMCP `Models` registry |
| Multi-agent patterns | QMCP Topologies |
| Structured output | PydanticAI `output_type` |
| Token limits | PydanticAI `UsageLimits` |
| Persistent memory | QMCP `MemoryMixin` |
## Architecture
```
┌─────────────────────────────────────────────────────────────┐
│ Your Application │
├─────────────────────────────────────────────────────────────┤
│ │
│ ┌─────────────────────┐ ┌─────────────────────┐ │
│ │ PydanticAI Agent │────▶│ QMCP Topologies │ │
│ │ (Runtime) │ │ (Orchestration) │ │
│ └─────────────────────┘ └─────────────────────┘ │
│ │ │ │
│ ▼ ▼ │
│ ┌─────────────────────────────────────────────────────┐ │
│ │ QMCP Model Registry │ │
│ │ (Pricing, Limits, Capabilities, Fallbacks) │ │
│ └─────────────────────────────────────────────────────┘ │
│ │ │
│ ▼ │
│ ┌─────────────────────────────────────────────────────┐ │
│ │ QMCP MCP Server (FastAPI) │ │
│ │ (Audit Trail, HITL, Metrics) │ │
│ └─────────────────────────────────────────────────────┘ │
│ │ │
│ ▼ │
│ ┌─────────────────────────────────────────────────────┐ │
│ │ QMCPToolset (PydanticAI Adapter) │ │
│ │ (Connect agents to QMCP server) │ │
│ └─────────────────────────────────────────────────────┘ │
│ │
└─────────────────────────────────────────────────────────────┘
```
## Future Integrations
Planned integrations:
| Integration | Purpose | Status |
|-------------|---------|--------|
| LangChain | Alternative agent framework | Planned |
| Instructor | Structured extraction | Considered |
| Marvin | AI functions | Considered |
