Description

n8n-nodes-mongodb-vector-search

High-performance n8n community node package for MongoDB Atlas Vector Search, Schema-Aware AI Agent Searching, standard queries, and custom JSON aggregation pipelines.

It provides three dedicated nodes designed for both standard n8n workflows and LangChain AI Agents:

1. 🔍 MongoDB Vector Search (Standalone Node): Fast vector search, find, and custom aggregations for regular workflow pipelines.
2. 🤖 MongoDB Vector Search (AI) (AI Store / Tool Node): Connects MongoDB Atlas Vector Search directly to n8n AI Agents as a Vector Store or Tool.
3. 🧠 MongoDB AI Search (Smart AI Search Node): A schema-aware AI Agent tool for searching MongoDB collections without embeddings, featuring automatic schema analysis, pre/post filters, and context window protection.

—

Installation

To install this community node package in your n8n instance:

1. Go to Settings > Community Nodes.
2. Click Install a new node.
3. Enter the npm package name: n8n-nodes-mongodb-vector-search.
4. Agree to the terms and click Install.

—

Included Nodes & Usage Guide

1. 🧠 MongoDB AI Search (mongoDbAiSearch)

Designed specifically for n8n AI Agents to search, filter, and inspect MongoDB collections without requiring vector embeddings.

#### Features

• Automatic Schema Discovery: On initialization, the node samples documents to discover field names, data types, and distinct sample values (e.g. paymentmethod: ["paypal", "creditcard"]).
• LLM Schema Prompt: Injecting discovered schema details into the AI Agent tool prompt so the LLM knows exact field names and categorical values without guessing.
• Field Privacy & Projection Mode:

– Return All Fields: Emits all document attributes.
– Include Only Specified Fields: Exposes only whitelisted fields.
– Exclude Specified Fields: Hides sensitive fields (e.g. passwordhash, secretkey).
– Exclude ID Field (id): Removes default id.

• Context Window Protection & Pagination: Supports buffer limits (limit, maxLimit) and outputs pagination metadata (totalCount, returnedCount, skip, hasMore, nextSkip).

#### How to Connect in n8n

• Output Port (Tool): Connect MongoDB AI Search‘s Tool output port directly to the AI Agent node’s Tools input port.

—

2. 🤖 MongoDB Vector Search (AI) (mongoDbVectorSearchVectorStore)

Connects MongoDB Atlas Vector Search to n8n AI Agents as a Vector Store or Tool.

#### Features

• Dual Connection Ports: Connects to AI Agents via Vector Store port or Tool port.
• Pre-Filtering ($vectorSearch.filter): Evaluated inside the vector search index stage before computing distance.
• Post-Filtering ($match): Evaluated after vector candidate retrieval on any document field.
• Dynamic Structured Tool (Zod Schema): Uses DynamicStructuredTool with a valid type: "object" Zod schema, ensuring 100% compatibility with n8n AI Agent tool calling engines.
• Automatic Key-Value Formatting: If documents lack a designated text field, all fields are formatted into readable key-value pairs for the AI Agent.

#### How to Connect in n8n
1. Embedding Model Input: Connect Embeddings Ollama or Embeddings OpenAI to the node’s Embedding Model input port (bottom).
2. AI Agent Output: Connect the node’s Vector Store or Tool output port (top) to the AI Agent node’s Vector Store or Tools input port.

—

3. 🔍 MongoDB Vector Search (Standalone) (mongoDbVectorSearch)

The clean standalone node for standard n8n workflows without AI sub-node handles.

#### Operations

• Vector Search: Queries Atlas Vector Search indexes using direct vector arrays or connected embedding models.
• Find (Normal Search): Standard MongoDB query filter (collection.find()).
• Custom Search: Execute custom queries or multi-stage $aggregate pipelines.

—

Features & Optimizations

• Connection Pooling: Global client cache reuses active MongoClient connections across executions for fast, low-latency queries.
• Extended JSON (EJSON): Supports {"$oid": "..."} and {"$date": "..."} data types natively.
• Dark & Light Mode Icons: Native high-resolution SVG iconography for n8n UI themes.

—

License

MIT