Skip to main content
Release Date: 2025-10-13 Breaking Changes: None (Pydantic V2 backward compatible)

​
Overview

Version 2.3.0 completes the compliance storage backend infrastructure, migrates to Pydantic V2, and significantly enhances type safety across the codebase.

​
Key Features

​
🗄️ Compliance Storage Backend

Complete pluggable storage architecture with 5 abstract interfaces and in-memory implementations. New Interfaces (src/mcp_server_langgraph/core/compliance/storage.py - 735 lines):
  • UserProfileStore - CRUD operations for user profiles
  • ConversationStore - Manage conversations with archival
  • PreferencesStore - User preferences key-value storage
  • AuditLogStore - Audit logging with date filtering
  • ConsentStore - GDPR consent tracking
In-Memory Implementations:
  • For development and testing without external dependencies
  • Production-ready interface for database backends
Benefits:
  • Pluggable architecture
  • Type safety with Pydantic models
  • GDPR compliant

​
🔒 Enhanced Type Safety

Coverage Increase: 27% → 64% Strict mypy typing enabled for additional modules:
  • mcp_server_langgraph.auth.*
  • mcp_server_langgraph.llm.*
  • mcp_server_langgraph.core.agent
Settings: disallow_untyped_calls = true, strict = true Impact: Catches type errors at development time, prevents runtime bugs

​
📋 Pydantic V2 Migration

Fixed Deprecations:
  • Replaced class Config with model_config = ConfigDict(...)
  • Replaced deprecated regex parameter with pattern in Query validators
Files Updated:
  • data_export.py: UserDataExport model
  • gdpr.py: UserProfileUpdate, ConsentRecord, ConsentResponse models
Impact: Eliminates 5 Pydantic deprecation warnings, future-proofs for Pydantic V3

​
📚 Error Handling Strategy Documentation

Created docs/architecture/adr-0006-error-handling-strategy (600+ lines):
  • 5 error categories (Client 4xx, Server 5xx)
  • Layered propagation pattern (4 layers)
  • Consistent JSON response format
  • Retry strategies (LLM: 3x, OpenFGA: 2x, Redis: 2x)
  • 40+ standardized error codes
  • OpenTelemetry integration

​
Files Modified

  1. storage.py (NEW) - 735 lines
    • 5 abstract interfaces
    • 6 Pydantic models
    • 5 in-memory implementations
  2. data_export.py (MODIFIED)
    • Integrated with UserProfileStore
    • Integrated with ConversationStore, PreferencesStore
    • Pydantic V2 migration
  3. retention.py (MODIFIED)
    • Cleanup inactive sessions integration
  4. session.py (MODIFIED)
    • Added get_inactive_sessions()
    • Added delete_inactive_sessions()
  1. pyproject.toml
    • Strict mypy for auth/llm/agent modules
  2. .flake8 (NEW)
    • Explicit configuration (47 lines)
    • Max line length: 127 (aligned with Black)
    • Google docstring convention

​
Dependency Management

​
New Documentation

  • docs/DEPENDENCY_MANAGEMENT.md (580 lines)
  • scripts/dependency-audit.sh (320 lines)
4-Phase Update Strategy:
  • Critical security: 48h SLA
  • Major versions: 2-4 weeks
  • Minor versions: 1-2 weeks
  • Patches: 1 month
Monthly Audit Script:
  • Outdated packages scan
  • Security vulnerability scan (pip-audit)
  • License compliance (pip-licenses)
  • Dependency conflict detection

​
Upgrade Guide

​
From v2.2.0

## 1. Update dependencies
git pull origin main
uv sync

## 2. Verify Pydantic V2
python -c "import pydantic; print(pydantic.__version__)"
## Expected: 2.x

## 3. Run type checks
mypy src/
## Should pass with new strict typing

## 4. Test compliance storage
pytest tests/test_storage.py -v

​
For Custom Extensions

If you’ve extended Pydantic models: 
## Before (v1.x)
class MyModel(BaseModel):
    class Config:
        arbitrary_types_allowed = True

## After (v2.x)
from pydantic import ConfigDict

class MyModel(BaseModel):
    model_config = ConfigDict(arbitrary_types_allowed=True)

​
What’s Next

​
v2.4.0

  • LangGraph 0.6.10 upgrade
  • 100% test pass rate
  • 21 comprehensive ADRs