MongoDB MCP 서버 사용 예시

Ran ``atlas-list-orgs``
Ran ``atlas-list-projects``
Here are your MongoDB Atlas organizations and projects:
**Organizations:**
TestOrganization
Organization ID: XXX
**Projects:**
TestProject
Project ID: XXX
Organization: TestOrganization
Created: August 7, 2020 at 11:48:55 AM
You have one organization containing one project. This
project contains your TestCluster that we're currently
connected to.

I'll create a new Atlas project, free cluster, and set up
all the necessary access details for you.
Ran ``atlas-create-project``
Great! I've created a new project named "myNewProject"
in your MongoDB Atlas organization. Now let's list the
projects to get the project ID for the new project:
Ran ``atlas-list-projects``
Now I'll create a free cluster named "myNewCluster" in the
new project:
Ran ``atlas-create-free-cluster``
Now, I'll add your current IP address to the access list:
Ran ``atlas-create-access-list``
Now let's create a database user named "myNewUser" with read
and write access:
Ran ``atlas-create-db-user``
Perfect! Everything has been set up successfully:
**New Project Information:**
Project Name: myNewProject
Project ID: XXX
**New Cluster Information:**
Cluster Name: myNewCluster
Region: US East (US_EAST_1)
**Database User:**
Username: myNewUser
Password: XXX
Access: Read and write to any database
**Connection String:**
mongodb+srv://myNewUser:XXX@myNewCluster.mongodb.net/
Your current IP address has been added to the access list,
so you should be able to connect to the cluster from your
current location. The cluster will take a few minutes to
fully provision and be ready for connections.

Ran ``atlas-get-regions``
Here are some of the supported Atlas regions on GCP:
CENTRAL_US — Iowa, USA
US_EAST_4 — Northern Virginia, USA
WESTERN_EUROPE — Belgium
EUROPE_WEST_3 — Frankfurt, Germany
EUROPE_NORTH_1 — Finland
ASIA_SOUTH_1 — Mumbai, India
AUSTRALIA_SOUTHEAST_1 — Sydney, Australia
Let me know if you'd like the full list or regions for
another provider.

I'll create a dedicated Atlas cluster on AWS
in US_EAST_1.
Ran ``atlas-create-cluster``
Cluster creation is underway. Here's a summary:
**Cluster Details:**
Name: myDedicatedCluster
Project ID: XXX
Provider: AWS
Regions: US_EAST_1
Instance Size: M10
Cluster Type: Replica Set
MongoDB Version: Latest
Backup: Snapshot
Compute Autoscaling: Enabled (M10–M30)
Disk Autoscaling: Enabled
The cluster is now being provisioned — it will take
a few minutes to reach IDLE state. You can use
``atlas-inspect-cluster`` to poll its status,
or connection strings will be available once it's ready.

I'll create a dedicated sharded Atlas cluster
on AWS in US_EAST_1.
Ran ``atlas-create-cluster``
Cluster creation initiated successfully. Here's
a summary:
**Cluster Details:**
Name: myShardedCluster
Project ID: XXX
Provider: AWS
Regions: US_EAST_1
Cluster Type: Sharded
Instance Size: M30
Backup: Continuous (point-in-time restore)
Compute Autoscaling: Enabled (M30–M50)
Disk Autoscaling: Enabled
The cluster is now being provisioned — it typically
takes a few minutes to reach IDLE state.
You can use ``atlas-inspect-cluster`` to poll its readiness,
and connection strings will be available once it's IDLE.

I'll create a dedicated Atlas cluster on AWS across
US_EAST_1, US_EAST_2, and EU_WEST_1.
Ran ``atlas-create-cluster``
Cluster creation is underway. Here's a summary:
**Cluster Details:**
Name: myMultiRegionCluster
Project ID: XXX
Provider: AWS
Regions: US_EAST_1 (primary), US_EAST_2, EU_WEST_1
Instance Size: M10
Cluster Type: Replica Set
Electable Nodes: 2 / 2 / 1
Backup: Snapshot
Compute Autoscaling: Enabled (M10–M30)
Disk Autoscaling: Enabled
The cluster is now being provisioned — it will take a
few minutes to reach IDLE state. You can use
``atlas-inspect-cluster`` to poll its status.

I'll create a dedicated Atlas cluster on AWS in
US_EAST_1 using your AWS KMS customer-managed key.
Ran ``atlas-create-cluster``
Cluster creation is underway. Here's a summary:
**Cluster Details:**
Name: myEncryptedCluster
Project ID: XXX
Provider: AWS
Regions: US_EAST_1
Instance Size: M10
Cluster Type: Replica Set
Encryption at Rest Provider: AWS
Backup: Snapshot
Compute Autoscaling: Enabled (M10–M30)
Disk Autoscaling: Enabled
The cluster is now being provisioned — it will take a
few minutes to reach IDLE state. You can use
``atlas-inspect-cluster`` to poll its status.

I'll upgrade myFreeCluster to Flex tier.
First, let me confirm the cluster details:
Ran ``atlas-inspect-cluster``
The cluster is a Free tier cluster eligible
for upgrade.
Now I'll perform the upgrade:
Ran ``atlas-upgrade-cluster``
The upgrade request was submitted successfully.
**Cluster Upgrade Details:**
Cluster Name: myFreeCluster
Previous Tier: Free
New Tier: Flex
Region: US East (US_EAST_1)
Provider: AWS
The upgrade may take a few minutes to complete.

I'll upgrade myFlexCluster to M10 dedicated
tier on AWS in US East.
First, let me confirm the cluster details:
Ran ``atlas-inspect-cluster``
The cluster myFlexCluster is a Flex cluster
eligible for upgrade to M10.
Now I'll perform the upgrade with the specified
provider and region:
Ran ``atlas-upgrade-cluster``
The upgrade request was submitted successfully.
**Cluster Upgrade Details:**
Cluster Name: myFlexCluster
Previous Tier: Flex
New Tier: M10 (Dedicated)
Cluster Type: Replica Set
Node Count: 3
Region: US East (US_EAST_1)
Provider: AWS
The upgrade may take a few minutes to complete.

I'll scale myDedicatedCluster to M30 with compute
autoscaling between M20 and M50.
First, let me confirm the cluster details:
Ran ``atlas-inspect-cluster``
The cluster myDedicatedCluster is an M10
dedicated cluster.
Now I'll scale the cluster:
Ran ``atlas-upgrade-cluster``
The scaling request was submitted successfully.
**Cluster Scaling Details:**
Cluster Name: myDedicatedCluster
Previous Tier: M10
New Tier: M30
Compute Autoscaling: Enabled (M20–M50)
The change may take a few minutes to complete.
You can use ``atlas-inspect-cluster`` to poll its
status.

I'll load the schema for the cluster pause tool first.
Pausing cluster test-cluster-1 now.
Ran ``atlas-pause-resume-cluster``
test-cluster-1 has been paused. It will not incur
compute costs while paused.

Resuming cluster test-cluster-1 in project XXX.
Ran ``atlas-pause-resume-cluster``
The resume request for test-cluster-1 in project XXX has
been submitted successfully. The cluster won't be
immediately available — use the inspect cluster tool to
poll its state until it shows IDLE.

Ran ``list-databases``
Here are the databases in your Atlas cluster "TestCluster":
**Sample Databases:**
sample_mflix - 138.4 MB (Movie database with films, theaters, users, comments)
sample_training - 53.6 MB (Training data for learning MongoDB)
sample_airbnb - 55.3 MB (Airbnb listings and reviews data)
sample_analytics - 10.0 MB (Analytics and customer data)
sample_restaurants - 6.8 MB (Restaurant and review data)
sample_weatherdata - 2.7 MB (Weather observation data)
sample_geospatial - 1.3 MB (Geospatial data examples)
sample_supplies - 1.1 MB (Supply chain data)
sample_guides - 41 KB (Getting started guides)
**System Databases:**
local - 4.1 GB (MongoDB internal database)
admin - 365 KB (Administrative database)
Your cluster contains MongoDB's sample datasets, which are
great for learning and testing. The largest databases are
sample_mflix (movies), sample_training, and sample_airbnb.
Would you like me to explore any specific database or show
you the collections within one of them?

Ran ``list-collections``
Here are the collections in the sample_mflix database:
**Core Collections:**
movies - Main collection containing movie data
(titles, genres, cast, ratings, etc.)
users - User profiles and account information
comments - User comments and reviews on movies
theaters - Movie theater locations and details
sessions - User session data
**Additional Collections:**
embedded_movies - Movies collection with embedded/denormalized data structure
The movies collection contains detailed information
about films including plot, cast, directors, ratings,
release dates, and genres. The comments collection links to
movies and users to store user reviews and discussions.
Would you like me to explore any specific collection in more
detail, such as showing sample documents or collection
statistics?

I'll help you analyze the Atlas movies collection focusing
on thrillers. First, let me check what databases are
available in your connected cluster.
Ran ``list-databases``
Perfect! I can see you have the sample_mflix database which
contains the movies collection. Let me check the collections
in that database:
Ran ``list-collections``
Great! Now let me analyze the movies collection to find
thrillers and provide a summary. First, let me check the
schema to understand how genres are stored:
Ran ``collection-schema``
Perfect! I can see that genres are stored as an array of
strings. Now let me get a count of thriller movies and then
perform an aggregation to summarize them:
Ran ``count``
Now let me create a comprehensive summary of thriller movies
using aggregation:
Ran ``aggregate``
I see there are some data quality issues. Let me get better
data by filtering for movies with numeric ratings:
Ran ``aggregate``
Now let me get some more summary statistics by decade:
Ran ``aggregate``
Here's a comprehensive analysis of thriller movies in your
Atlas sample_mflix collection:
**Overall Statistics**
Total Thriller Movies: 2,454 films
Time Span: 1924 to 2011
Average Rating: 6.3/10 (IMDb)
Average Runtime: 105 minutes
Total IMDb Votes: 85.8 million across all thrillers
**Top-Rated Thriller Movies**
Drishyam (2013) - 8.9/10 - Indian thriller directed by Jeethu Joseph
Tinker Tailor Soldier Spy (1979) - 8.7/10 - British spy thriller
Rear Window (1954) - 8.6/10 - Alfred Hitchcock classic
The Silence of the Lambs (1991) - 8.6/10 - Psychological thriller
Se7en (1995) - 8.6/10 - David Fincher crime thriller
The Usual Suspects (1995) - 8.6/10 - Neo-noir mystery
Léon: The Professional (1994) - 8.6/10 - Action thriller
Psycho (1960) - 8.6/10 - Another Hitchcock masterpiece
Memento (2000) - 8.5/10 - Christopher Nolan psychological thriller
**Trends by Decade**
1920s-1950s: Few thrillers but high quality (avg 7.2-7.8 rating)
1960s-1970s: Growing popularity, still maintaining quality
1980s: Significant growth (200 films), quality decline begins
1990s: Peak production era (417 films), avg 6.3 rating
2000s: Highest volume (854 films), avg 6.24 rating
2010s: 730 films, lowest average rating (6.12)
The data shows that while thriller production increased
dramatically from the 1980s onward, the average quality (as
measured by IMDb ratings) has generally declined, suggesting
the genre became more commercialized with higher volume but
potentially lower artistic standards in recent decades.

I'll add a new document to the movies_copy collection. Let
me create a sample movie document with the same structure as
the existing movies:
Ran ``insert-many``
Perfect! I've successfully added a new movie document to the
movies_copy collection. Here's what I added:
**New Movie Added:**
Title: "The AI Thriller"
Genres: Thriller, Sci-Fi, Drama
Year: 2024
Runtime: 118 minutes
IMDb Rating: 7.8/10 (125,000 votes)
Cast: John Smith, Sarah Johnson, Michael Chen
Director: Alex Rodriguez
Rating: PG-13
The document was inserted with ID: 684b15fb5084ddef1f265e8f
Let me verify the collection now has one additional
document:
Ran ``count``
Confirmed! The movies_copy collection now contains 21,350
documents (up from 21,349), including your new thriller
movie about AI consciousness.

I'll create a new collection for storing movie purchases
data with geospatial and timeseries fields. Let me first
create the collection and then add some sample documents
with the appropriate structure.
Ran ``create-collection``
Now I'll insert sample documents that include geospatial
data (customer location, theater location) and timeseries
data (purchase timestamp, viewing timestamp):
Ran ``insert-many``
Perfect! I've created the movie_purchases collection with
sample data that includes both geospatial and timeseries
fields. Here's what the collection contains:
**Collection Structure**
**Geospatial Fields:**
customer_location: GeoJSON Point with customer's coordinates
theater_location: GeoJSON Point with theater coordinates
customer_address: Human-readable address information
**Timeseries Fields:**
purchase_timestamp: When the purchase was made
viewing_timestamp: When the movie was actually watched
viewing_duration_minutes: How long the customer watched
completion_percentage: Percentage of movie completed
**Sample Data Added:**
Purchase 1: Customer in San Francisco bought "The Shawshank Redemption" rental for $12.99
Purchase 2: Customer in New York purchased "Pulp Fiction" for $15.99
Purchase 3: Customer in Los Angeles rented "Se7en" for $9.99
Now let me create a geospatial index to optimize
location-based queries:
Ran ``create-index``
Excellent! The movie_purchases collection is now ready with:
- Geospatial capabilities - 2dsphere index for location-based queries
- Timeseries optimization - Index on purchase timestamps
- Sample data - 3 purchase records across different US cities
This collection can now support queries like:
- Finding purchases within a certain radius of a location
- Analyzing purchase patterns over time
- Correlating viewing behavior with geographic regions
- Time-based analytics and trending

I'll create a copy of the movies collection named
movies_copy. I'll use an aggregation pipeline with $out to
copy all documents from the movies collection to a new
movies_copy collection.
Ran ``aggregate``
Perfect! I've successfully created a copy of the movies
collection named movies_copy. Let me verify the copy was
created correctly by checking the document count:
Ran ``count``
Found 21349 documents in the collection "movies_copy"
Excellent! The movies_copy collection has been successfully
created with all 21,349 documents from the original movies
collection. The copy operation completed successfully and
you now have an identical backup of your movies collection.

I'll help you summarize and export the contents of the movies
collection from your connected MongoDB Atlas cluster. Let's do
this in two steps:
1. First, I'll create a summary of the movies collection using an
   aggregation pipeline.
2. Then I'll export the results using the export tool.
Let's start by getting a summary of the movies collection:
Ran ``aggregate``
Now that we have a summary of the movies collection, let's create
a more focused dataset to export. Let's get the top 10
highest-rated movies:
Ran ``aggregate``
Now, let's export this data using the MCP Server export tool:
Ran ``export``
I've successfully summarized and exported the contents of the
movies collection from your MongoDB Atlas cluster. Here's a
breakdown of what I did:
**Movies Collection Summary**
**Collection Overview:**
The collection contains 21,349 movie documents
Time span: 1896 to 2016 (120 years of cinema history)
Average IMDb rating: 6.66 out of 10
Average runtime: ~104 minutes
**Top Genres:**
Drama (12,385 movies)
Comedy (6,532 movies)
Romance (3,318 movies)
Crime (2,457 movies)
Thriller (2,454 movies)
**Top Directors:**
Woody Allen (40 movies)
Martin Scorsese (32 movies)
Takashi Miike (31 movies)
Steven Spielberg (29 movies)
John Ford (29 movies)
**Top Languages:**
English (14,639 movies)
French (2,405 movies)
Spanish (1,600 movies)
German (1,377 movies)
Italian (1,228 movies)
**Highest Rated Movies:**
Band of Brothers (2001) - 9.6/10
Planet Earth (2006) - 9.5/10
The Civil War (1990) - 9.4/10
A Brave Heart: The Lizzie Velasquez Story (2015) - 9.4/10
Cosmos (1980) - 9.3/10
**Export Details**
The data has been exported to a JSON file in
relaxed format. You can access the exported file at:
/Users/XXX/.mongodb/mongodb-mcp/exports/68b8c10a6d0291128800be0b/sample_mflix.movies.68b8c2496d0291128800be0d.json
The export includes:
- Genre summary with counts and average ratings
- Runtime statistics (min, max, average)
- Year-by-decade summary with counts and ratings
- Top directors with film counts
- Language distribution
- Top 10 highest-rated movies
This export provides a comprehensive overview of the movies
collection that can be easily shared or imported into other tools
for further analysis or visualization.

I'll list all the local Atlas deployments for you.
Ran ``atlas-local-list-deployments``
Here are all your local Atlas deployments:
[
  {
    "name": "local7356",
    "state": "RUNNING",
    "mongoDbVersion": "8.2.1"
  }
]
You currently have one Atlas Local deployment named local7356
that is running MongoDB version 8.2.1.

I'll connect to your local Atlas deployment local7356 and then
list the databases.
Ran ``atlas-local-connect-deployment``
Now I'll list the databases in this deployment:
Ran ``list-databases``
Successfully connected to Atlas Local deployment local7356! Here
are the databases in this deployment:
Found 4 databases:
admin - 262,144 bytes (256 KB)
config - 237,568 bytes (~232 KB)
local - 1,163,264 bytes (~1.1 MB)
testdb - 438,272 bytes (~428 KB)
Your local Atlas deployment contains:
System databases: admin, config, and local (standard MongoDB
system databases)
User database: testdb (your custom database)
The local database has grown to about 1.1 MB, which is normal for
local operations and metadata storage. Would you like me to
explore any of these databases further, such as showing the
collections in testdb?

I'll help you create a new local Atlas deployment, connect to it,
create a collection with sample pizza orders data, and then
retrieve it. Let me break this down into steps:
Created 4 todos
Now let me start with the first task:
Starting (1/4) Create new local Atlas deployment
Ran ``atlas-local-create-deployment``
Great! Created deployment local5528. Now let me mark that as
completed and connect to it:
Starting (2/4) Connect to new deployment
Ran ``atlas-local-connect-deployment``
Perfect! Now let me create the pizza orders collection with sample
data:
Starting (3/4) Create pizza orders collection
Ran ``create-collection``
Now let me insert sample pizza orders data:
Great! Now let me retrieve and display the sample data:
Starting (4/4) Retrieve and display sample data
Ran ``find``
Completed (4/4) Retrieve and display sample data
Perfect! I've successfully completed all the tasks. Here's a
summary of what was accomplished:
Task Completed Successfully
1. Created New Local Atlas Deployment
Deployment Name: local5528
Container ID: XXX
Status: Running
2. Connected to the New Deployment
Successfully connected to deployment local5528
3. Created Pizza Orders Collection
Database: pizzarestaurant
Collection: orders
Documents Inserted: 5 pizza orders
4. Sample Pizza Orders Data Created
The collection now contains 5 realistic pizza orders with the
following details:
Order ORD-001 - John Smith (New York)
Margherita Pizza (Large) + Pepperoni Pizza (Medium)
Total: $34.98 | Status: Completed | Delivery
Order ORD-002 - Sarah Johnson (Brooklyn)
2x Hawaiian Pizza (Large) + Caesar Salad
Total: $50.97 | Status: Preparing | Pickup
Order ORD-003 - Mike Chen (Manhattan)
Meat Lovers Pizza (Extra Large) + 2x Garlic Bread + 3x Soda
Total: $40.95 | Status: Delivered | Delivery
Order ORD-004 - Emily Davis (Queens)
Veggie Supreme Pizza (Medium)
Total: $17.99 | Status: Ready | Pickup
Order ORD-005 - David Wilson (Staten Island)
BBQ Chicken Pizza (Large) + Buffalo Wings
Total: $35.98 | Status: Out for Delivery | Delivery
Data Structure Features:
Customer Information: Name, phone, email, complete address
Order Details: Order ID, date, delivery type, payment method, status
Items: Pizza types, sizes, quantities, prices, detailed toppings
Financial Data: Individual item prices and total amounts
Your new local Atlas deployment is now ready with a fully
functional pizza restaurant orders system!

I'll delete your local Atlas deployment local5528 for you.
Ran ``atlas-local-delete-deployment``

Ran ``atlas-get-performance-advisor``
Based on the performance advisor analysis of your MongoDB Atlas cluster "ClusterM10", here are the specific index recommendations:
**Index to CREATE:**
**High Priority - Create Index on sample_db.test collection:**
- **Index**: { "z": 1 }
- **Weight**: 1,020,000 bytes (≈ 1.02 MB saved per query)
- **Impact**: This index will significantly improve performance for queries filtering on the z field
- **Reason**: The slow query logs show a query {"z": 10} that performed a collection scan (COLLSCAN) and examined 10,001 documents to return just 1 result
**Indexes to DROP:**
**No Drop Recommendations Found**
The performance advisor found no indexes that should be dropped at this time.
**Additional Performance Insights:**
**Schema Recommendations:**
- **Remove Unnecessary Indexes**: You have collections with more than 30 indexes, which can impact write performance
- **Reduce Document Size**: Some documents exceed 2MB, which can cause performance issues
- **Avoid Unbounded Arrays**: Some collections have arrays with over 10,000 entries
**Slow Query Analysis:**
The slow query logs show that most slow queries are from MongoDB's internal automation agent running $collStats operations, which is normal. However, there's one user query that would benefit significantly from the suggested index:
- **Query**: db.dbC.find({"z": 10})
- **Current Performance**: Collection scan examining 10,001 documents
- **Expected Improvement**: With the suggested index, this should become an index scan examining only the matching documents
**Recommended Action:**
Create the index { "z": 1 } on the sample_db.test collection. This single index creation could save approximately 1MB of disk reads per query execution for queries filtering on the z field.

Ran ``atlas-get-performance-advisor``
Based on the performance advisor analysis of your MongoDB Atlas cluster, here are the specific schema improvements recommended:
**Critical Schema Issues Found:**
**1. Remove Unnecessary Indexes**
**Collections Affected:**
- SchemaAdvisor.Call to Action: Remove unnecessary indexes
- SchemaAdvisor.Call to Action: Remove unnecessary indexes 2
**Issue**: Both collections have more than 30 indexes each
**Impact:**
- Slower write operations
- Increased storage overhead
- Longer index maintenance time
- Reduced memory efficiency
**Recommendation:**
- Audit all indexes in these collections
- Remove indexes that are not actively used by your queries
- Keep only indexes that support your application's query patterns
- Consider compound indexes instead of multiple single-field indexes
**2. Reduce Document Size**
**Collections Affected:**
- SchemaAdvisor.Call to Action: Reduce the size of documents
- SchemaAdvisor.Call to Action: Reduce the size of documents 2
**Issue**: Documents larger than 2MB found
**Impact:**
- Documents cannot be stored in MongoDB (16MB limit)
- Performance degradation for large documents
- Increased memory usage
- Slower network transfer
**Recommendations:**
- Split large documents into smaller, related documents
- Use GridFS for storing large binary data
- Normalize data by moving large arrays or objects to separate collections
- Archive old data to reduce document size
- Compress data where appropriate
**3. Avoid Unbounded Arrays**
**Collections Affected:**
- SchemaAdvisor.Call to Action: Avoid using unbounded arrays in documents
**Issue**: Arrays with over 10,000 entries detected
**Impact:**
- Poor query performance on array elements
- Memory consumption issues
- Index inefficiency
- Difficult to maintain and update
**Recommendations:**
- Move large arrays to separate collections with references
- Implement pagination for array access
- Use embedded documents with limits (e.g., max 100 items)
- Consider denormalization for frequently accessed array elements

I'll help you create a vector search index on the MongoDB collection. Let me first check the available MongoDB tools and then create the index.
Now I'll create the vector search index:
  Ran ``create-index``
Vector search index created successfully!
The index embeddings_vector_index has been created on the sample_db.products collection for the embeddings field with:
1024 dimensions
Dot product similarity
Note that vector search indexes can take some time to build. You can check the index status using the collection-indexes tool.

Ran ``collection-indexes``
The sample_db.products collection has 1 vector search index:
embeddings_vector_index
Type: vectorSearch
Status: READY ✓
Queryable: Yes
Field: embeddings
Dimensions: 2048
Similarity: dotProduct
Quantization: none
The index is ready and can be used for vector search queries.

First, let me check what indexes exist on this collection to identify the vector search index.
I can see there's a vector search index named embeddings_vector_index on the collection.
Now I need to activate the index management tools to drop it.
  Ran ``drop-index``
Done! The vector search index embeddings_vector_index has been successfully dropped from the sample_db.products collection.

I'll insert these documents with embeddings for their descriptions.
Let me activate the necessary tools and generate embeddings.
  Ran ``insert-many``
Successfully inserted 5 documents into sample_db.products:
Headphones - Premium wireless noise-canceling headphones with 30-hour battery
Earbuds - True wireless earbuds with active noise cancellation
Monitor - 24-inch 4K monitor with 1ms response time
Keyboard - Mechanical keyboard with customizable RGB lighting
Mouse - Wireless ergonomic mouse with programmable buttons
The documents have been inserted with their descriptions embedded in the embeddings field.
When you perform vector searches using the $vectorSearch aggregation stage, the MCP server
will automatically generate the proper vector embeddings from the text for comparison.

I'll search for products related to listening to music using vector search.
Ran ``aggregate``
Products for Listening to Music
1. Earbuds (Score: 0.77)
   True wireless earbuds with active noise cancellation
2. Headphones (Score: 0.77)
   Premium wireless noise-canceling headphones with 30-hour battery
These are the two most relevant products for listening to music in your database.
Both scored highly on the semantic search and are specifically designed for audio listening,
with the earbuds and headphones being the clear top matches.

I'll help you create a lexical search index on the MongoDB collection. Let me first check the available MongoDB tools and then create the index.
Now I'll create the lexical search index:
  Ran ``create-index``
Lexical search index created successfully!
The index description_search_index has been created on the sample_db.products collection for the description field.
Note that lexical search indexes can take some time to build. You can check the index status using the collection-indexes tool.

Ran ``collection-indexes``
The sample_db.products collection has 1 lexical search index:
description_search_index
Type: search
Status: READY ✓
Queryable: Yes
Mapped field: description
The index is ready and can be used for lexical search queries.

First, let me check what indexes exist on this collection to identify the lexical search index.
I can see there's a lexical search index named description_search_index on the collection.
Now I need to activate the index management tools to drop it.
  Ran ``drop-index``
Done! The lexical search index description_search_index has been successfully dropped from the sample_db.products collection.

I'll search for products that mention wireless using lexical search.
Ran ``aggregate``
Products Matching "wireless"
1. Headphones (Score: 1.42)
   Premium wireless noise-canceling headphones with 30-hour battery
2. Mouse (Score: 1.15)
   Wireless ergonomic mouse with programmable buttons
These are the top matches for products whose description mentions "wireless."

I'll get the count of products that mention wireless using lexical search metadata.
Ran ``aggregate``
5 products in sample_db.products have "wireless" in the description.