INTPYTHON-809 - Update to langchain 1.0 and langgraph 1.0

[e-palmisano]

INTPYTHON-809 - Migrate langchain-mongodb to LangChain 1.0

Issue Key

Summary

This PR migrates the langchain-mongodb package to be compatible with LangChain 1.0+. The main focus is on removing dependencies on legacy packages (langchain-classic) where possible and updating retrievers to use the modern
langchain_core APIs. This includes:

• Refactoring MongoDBAtlasParentDocumentRetriever to extend BaseRetriever directly from langchain_core instead of the legacy ParentDocumentRetriever
• Fixing critical bugs in the add_documents() method discovered during migration
• Updating MongoDBAtlasSelfQueryRetriever to explicitly use langchain-classic for components not yet available in LangChain 1.0
• Removing the langchain < 1.0 version pin to allow users to upgrade to LangChain 1.0+
• Updating dependencies for compatibility with langgraph-checkpoint>=3.0.1 and aiohttp>=3.13.2

Changes in this PR

1. MongoDBAtlasParentDocumentRetriever Refactoring

File: libs/langchain-mongodb/langchain_mongodb/retrievers/parent_document.py

• Removed dependency on langchain-classic: Changed base class from ParentDocumentRetriever (legacy) to BaseRetriever from langchain_core
• Added child_splitter field: Required for splitting parent documents into chunks
• Added search_kwargs field: Enables passing additional search parameters to vector search
• Implemented add_documents() method: Complete implementation for adding parent documents and child chunks to the vectorstore

Critical Bug Fixes in add_documents():

• Fixed call to non-existent docstore.add_documents() → changed to docstore.mset() which is the correct API for MongoDBDocStore
• Fixed indentation bug where vectorstore.add_documents() and docstore.mset() were called inside the document loop instead of after processing all documents
• Added proper key-value pair formatting for mset(): [(id, Document), ...]

Updated Docstrings:

• Removed references to legacy ParentDocumentRetriever and MultiVectorRetriever classes
• Added documentation about LangChain 1.0+ compatibility
• Fixed example code to use correct class name MongoDBAtlasParentDocumentRetriever

2. MongoDBAtlasSelfQueryRetriever Dependencies

File: libs/langchain-mongodb/langchain_mongodb/retrievers/self_querying.py

• Updated imports to use langchain-classic for AttributeInfo and SelfQueryRetriever
• These classes are still in langchain-classic in LangChain 1.0 as they are considered legacy components
• Added documentation noting the langchain-classic dependency requirement

3. Documentation Updates

File: libs/langchain-mongodb/CHANGELOG.md

• Added version 0.8.0 entry with detailed breakdown of:
  • Breaking changes (LangChain 1.0+ requirement, API changes)
  • Bug fixes (docstore and indentation issues)
  • Enhancements (new fields, updated docs)

4. Dependency Updates

Files: pyproject.toml, uv.lock

• Removed pin langchain < 1.0
• Updated to require langchain >= 1.0
• Added/updated dependencies:
  • langgraph-checkpoint >= 3.0.1
  • aiohttp >= 3.13.2

Test Plan

Unit Tests

• Existing tests: All existing unit tests for MongoDBAtlasParentDocumentRetriever pass with the refactored implementation
• Key tested scenarios:
  • Document splitting and chunk creation
  • Metadata propagation (doc_id field)
  • Vector search with parent document lookup
  • Docstore operations using mset()

Integration Tests

• Verified that MongoDBAtlasParentDocumentRetriever.from_connection_string() correctly initializes the retriever
• Tested add_documents() with multiple parent documents
• Verified retrieval returns complete parent documents (not chunks)
• Confirmed MongoDBAtlasSelfQueryRetriever works with langchain-classic installed

Manual Testing

• Created a sample script using MongoDBAtlasParentDocumentRetriever with LangChain 1.0
• Verified document indexing and retrieval workflow end-to-end
• Confirmed no import errors or deprecation warnings with LangChain 1.0+

Compatibility Testing

• Tested with langchain==1.0.0
• Tested with langchain-core==1.0.0
• Verified langchain-classic is correctly used only where necessary

Checklist

Checklist for Author

• Did you update the changelog (if necessary)?
• Is the intention of the code captured in relevant tests?
• If there are new TODOs, has a related JIRA ticket been created?
• Has a MongoDB Employee run the patch build of this PR?

Checklist for Reviewer {@primary_reviewer}

• Does the title of the PR reference a JIRA Ticket?
• Do you fully understand the implementation? (Would you be comfortable explaining how this code works to someone else?)
• Have you checked for spelling & grammar errors?
• Is all relevant documentation (README or docstring) updated?

Focus Areas for Reviewer

1. add_documents() implementation in parent_document.py: This method was completely rewritten to fix critical bugs. Please verify:

   • The logic for splitting documents into chunks is correct
   • The use of docstore.mset() with key-value pairs is appropriate
   • The operations are correctly placed outside the document loop

2. Breaking changes: This PR introduces breaking changes for users:

   • MongoDBAtlasParentDocumentRetriever no longer requires langchain-classic but has a different API surface
   • MongoDBAtlasSelfQueryRetriever now requires explicit installation of langchain-classic
   • Verify the CHANGELOG adequately documents these changes

3. LangChain 1.0 compatibility:

   • Ensure all imports from langchain_core are correct
   • Verify no deprecated APIs are being used
   • Check that the migration path is clear for users

4. Backward compatibility concerns:

   • Are there any users who might be affected by the removal of the langchain < 1.0 pin?
   • Should we provide a migration guide in the documentation?

[blink1073]

Hi @e-palmisano, thank you for creating this PR. We had to prioritize getting security fixes for the langgraph packages, so there are now merge conflicts in this PR. I think we can make langchain_classic a test dependency, and an optional dependency for the self-querying retriever. It could raise an informative error on class instantiation if the import was not found.

[blink1073]

Closing in favor of #271. The 0.8 release is now compatible with langchain>1.0.