MongoDB integration for LangChain

Warnings

• breaking LangChain's ecosystem split led to many integrations, including MongoDB, moving from `langchain` or `langchain-community` into dedicated packages like `langchain-mongodb`. Older import paths are deprecated or will result in `ModuleNotFoundError`.
  Fix: Ensure `langchain-mongodb` is installed: `pip install langchain-mongodb`. Update all imports to use `from langchain_mongodb...` instead of `from langchain...` or `from langchain_community...`.
• gotcha `MongoDBAtlasVectorSearch` requires a pre-configured Atlas Search Index (type 'Vector Search') on your MongoDB collection. If the `index_name` specified in your code does not exist or is misconfigured, operations will fail.
  Fix: Create a vector search index in the MongoDB Atlas UI for your target collection. Ensure the `index_name` parameter in `MongoDBAtlasVectorSearch` matches the name of your Atlas Vector Search index. Configure the index to use the correct embedding field and dimensions.
• gotcha All vector store operations (adding documents, performing similarity searches) require an instantiated embedding model. Forgetting to provide one or providing an incorrectly configured model will lead to errors.
  Fix: Pass a valid embedding model instance (e.g., `OpenAIEmbeddings(api_key="...")`, `HuggingFaceEmbeddings()`) to the `embedding` parameter of the `MongoDBAtlasVectorSearch` constructor.
• gotcha The MongoDB connection URI (`MONGODB_ATLAS_CLUSTER_URI`) must be correctly formatted, especially for Atlas clusters (e.g., `mongodb+srv://user:pass@cluster-name.mongodb.net/`). Incorrect protocols or missing credentials will result in connection failures.
  Fix: Verify your connection string directly from the MongoDB Atlas UI. Ensure it includes the correct protocol, host, and authentication credentials. For local setups, `mongodb://localhost:27017/` is common.

Install

• pip install langchain-mongodb pymongo Install core package and PyMongo
• pip install langchain-mongodb pymongo langchain-openai Install with OpenAI for embeddings

Imports

• MongoDBAtlasVectorSearch

  from langchain_mongodb.vectorstores import MongoDBAtlasVectorSearch

  MongoDB vector store integration moved to a dedicated package since LangChain's ecosystem split.
• MongoDBLoader

  from langchain_mongodb import MongoDBLoader

  Document loaders specific to MongoDB are now part of the langchain-mongodb package.
• MongoDBChatMessageHistory

  from langchain_mongodb.chat_message_histories import MongoDBChatMessageHistory

  Chat message history integrations are within the dedicated langchain-mongodb package.

Quickstart

This quickstart demonstrates how to use `MongoDBAtlasVectorSearch` to store and query documents. It connects to a MongoDB cluster (defaults to local if env vars aren't set), initializes a vector store with a placeholder embedding model, adds documents, and performs a similarity search. Remember to replace `DummyEmbeddings` with a real embedding model (e.g., `OpenAIEmbeddings`) for production use and create a vector search index in MongoDB Atlas.

import os
from pymongo import MongoClient
from langchain_mongodb.vectorstores import MongoDBAtlasVectorSearch

# NOTE: Replace DummyEmbeddings with a real embedding model (e.g., OpenAIEmbeddings)
# For a runnable example without extra API keys, we use a placeholder.
class DummyEmbeddings:
    def embed_documents(self, texts):
        # Return a list of fixed-size vectors for each text
        return [[0.1] * 1536 for _ in texts]
    def embed_query(self, text):
        # Return a fixed-size vector for a single query
        return [0.1] * 1536

# Environment variables for MongoDB connection
MONGODB_ATLAS_CLUSTER_URI = os.environ.get(
    "MONGODB_ATLAS_CLUSTER_URI", "mongodb://localhost:27017/"
)
MONGODB_DATABASE = os.environ.get("MONGODB_DATABASE", "langchain_db")
MONGODB_COLLECTION = os.environ.get("MONGODB_COLLECTION", "vector_collection")

# Initialize MongoDB client and collection
client = MongoClient(MONGODB_ATLAS_CLUSTER_URI)
collection = client[MONGODB_DATABASE][MONGODB_COLLECTION]

# Initialize embedding model (replace DummyEmbeddings with e.g., OpenAIEmbeddings)
# embeddings = OpenAIEmbeddings(openai_api_key=os.environ.get("OPENAI_API_KEY"))
embeddings = DummyEmbeddings()

# Initialize MongoDB Atlas Vector Search
# Ensure 'default' index exists in MongoDB Atlas on the specified collection
vector_search = MongoDBAtlasVectorSearch(
    collection=collection,
    embedding=embeddings,
    index_name="default", # The name of your Atlas Search Vector Index
)

# Add documents to the vector store
docs = [
    "The quick brown fox jumps over the lazy dog.",
    "A group of cats is called a clowder.",
    "Python is a high-level, interpreted programming language."
]
vector_search.add_texts(docs)
print(f"Added {len(docs)} documents to MongoDB Atlas Vector Search.")

# Perform a similarity search
query = "animals running"
results = vector_search.similarity_search(query, k=1)
print(f"Similarity search results for '{query}':")
for res in results:
    print(f"- {res.page_content}")

# Clean up (optional) - remove added documents
# collection.delete_many({"text": {"$in": docs}})
# print("Cleaned up documents.")