# Marqo AI MCP

> Marqo AI (Vector Search & Embeddings) lets you manage entire semantic search infrastructures through natural conversation. You can run dense similarity searches, upload and index new JSON documents instantly, or audit your vector indices without writing complex API calls. Gain full control over document lifecycle management—from creating bounded indexes to deleting specific vectors.

## Overview
- **Category:** ai-frontier
- **Price:** Free
- **Tags:** semantic-search, vector-embeddings, tensor-search, indexing, information-retrieval

## Description

Connecting Marqo AI to your agent lets you manage semantic search infrastructure entirely via chat. You don't need to write boilerplate code just to check what data exists or how relevant a concept is. Instead, you simply ask your agent questions like, 'Show me all the indexes we have,' or 'Find the best product match for lightweight running shoes.' This MCP handles everything: it executes complex tensor searches against your stored knowledge, writes fresh JSON records into your indices instantly, and helps you manage the whole index lifecycle by creating new search boundaries. When you're ready to scale this capability across multiple platforms, remember that Vinkius hosts this MCP, giving your agent access to thousands of tools in one place.

## Tools

### list_indexes
Lists all the available vector indexes within your Marqo instance so you know what collections exist.

### get_index_stats
Retrieves detailed configuration and operational statistics for a specific index.

### tensor_search
Performs deep semantic similarity searches using natural language queries against your indices.

### add_documents
Writes new structured documents into Marqo, making them instantly available for search.

### delete_documents
Physically removes specific documents from the index by referencing their unique IDs.

### create_index
Builds a new, isolated vector index with custom rules and constraints for a dedicated search purpose.

## Prompt Examples

**Prompt:** 
```
Semantic search in index 'products' for 'lightweight running shoes for trails'
```

**Response:** 
```
Executing tensor search… I've found 5 highly relevant products. Top match: 'Swift-Trail Runner' (Score: 0.89). Other results include specialized mountain gear and ultra-light hikers. Would you like to see the full JSON metadata for these matches?
```

**Prompt:** 
```
List all vector indexes in my Marqo instance
```

**Response:** 
```
I've identified 3 indexes: 'products' (Ecommerce catalog), 'support-docs' (Technical KB), and 'user-profiles' (Personalization data). Which one would you like to check the stats for?
```

**Prompt:** 
```
Add this document to the 'support-docs' index: {"title": "API Auth", "content": "Use Marqo-API-Key header"}
```

**Response:** 
```
Document added successfully to 'support-docs'. Marqo has vectorized the content using your configured embedding model. The document is now indexed and immediately available for semantic searches. Your new doc ID is 'marqo-12345'.
```

## Capabilities

### Perform semantic searches
Run natural language queries against your entire knowledge base to find highly relevant documents.

### Add new indexed data
Write fresh JSON records directly into your vector indices, making brand-new information immediately searchable by the agent.

### Manage index boundaries
Create explicitly defined vector indexes with custom rules and model settings for specific project needs.

### Audit index configurations
Retrieve detailed statistics, including document counts and embedding models, to check the health of your indices.

### Clean up old vectors
Delete specific documents or vectorized representations by targeting their unique IDs.

## Use Cases

### Updating a product catalog
A developer needs to update 50 new product descriptions in the vector store. Instead of writing a script, they simply ask their agent to use add_documents with the JSON data dump. The documents are indexed and available for search instantly.

### Diagnosing a stale index
The ML Engineer suspects an old index is holding garbage data. They first call list_indexes, then get_index_stats on the target index to verify document counts before running delete_documents to clean out outdated records.

### Building a feature store
A search architect wants a dedicated index just for user profiles. They use create_index first, setting up constraints, and then use add_documents repeatedly to populate it before testing with tensor_search.

### Retrieving context from multiple sources
The agent needs to find the best shoe recommendation. It uses tensor_search on the 'products' index but first uses get_index_stats to confirm that index is running the correct embedding model.

## Benefits

- Stop writing boilerplate code for basic checks. You can use list_indexes to see all available vector indices immediately, letting you know exactly what data sources your search needs.
- The agent handles the complex math behind tensor_search. Instead of feeding it a query vector, you just ask a question in plain English and get highly relevant results back.
- Keep your knowledge base clean using delete_documents. You target documents by ID to ensure that old or irrelevant vectors are physically removed from the index.
- Need a dedicated search silo? Use create_index to build a new, bounded vector index with specific model rules, keeping unrelated data separate and optimized.
- When you add_documents, your agent automatically handles embedding extraction. You just provide the JSON content; it becomes immediately searchable.

## How It Works

The bottom line is you get full control over complex vector database operations using simple conversation prompts.

1. Subscribe to this MCP and enter your Marqo API URL along with the necessary API Key.
2. Your agent connects these credentials, giving it immediate access to manage your vector search environment.
3. Start by asking your agent to list all available indexes or perform a semantic query from any MCP-compatible client.

## Frequently Asked Questions

**How does Marqo AI (Vector Search & Embeddings) MCP perform semantic searches?**
It uses the tensor_search tool to run dense similarity queries. You simply ask a question, and the agent handles turning that natural language into a vector query against your indexes.

**What should I use first when setting up Marqo AI (Vector Search & Embeddings)?**
Start by calling list_indexes. This action shows you every collection currently available on your instance, helping you map out your data landscape before running any queries.

**Can I update my knowledge base with Marqo AI (Vector Search & Embeddings) MCP?**
Yes, use the add_documents tool. You provide new JSON records to the agent, and it automatically processes them into your existing vector indices.

**Is there a way to isolate specific data sets in Marqo AI (Vector Search & Embeddings) MCP?**
You can use create_index. This tool builds an explicitly bounded, new vector index tailored for a very specific purpose or project.

**What if I find old documents that need removing in Marqo AI (Vector Search & Embeddings) MCP?**
Use delete_documents. You target the specific IDs of the vectors you want to remove, keeping your search index clean and highly relevant.