MongoDB Atlas Vector Search MCP. Query vectors and manage documents in one chat session.
Gemini Works with every AI agent you already use
…and any MCP-compatible client
Just plug in your AI agents and start using Vinkius.
MongoDB Atlas Vector Search gives you full control over high-performance vector search and operational data management directly from your AI agent.
Use it to execute sophisticated $vectorSearch queries, manage standard MQL documents (find/insert/delete), and provision custom search indices across your MongoDB Atlas cluster.
What your AI agents can do
Create index
Creates a standard embedding Search Index bound to specific dimensions in your collection.
Delete
Deletes documents from a target collection, using parsed MongoDB filters to specify the data set.
Find
Retrieves standard MongoDB documents by resolving specific query filters against your collections.
Execute advanced $vectorSearch queries using raw embedding vectors against your specified collections.
Use MQL filters to insert, find, or delete standard MongoDB documents in any target collection.
Build and configure Atlas Search indices by specifying custom dimensions and mapping definitions for your data.
Retrieve a list of all accessible data collections within the defined Atlas limits.
Run targeted MQL queries to fetch specific metadata or structural chunks, bypassing vector logic for rapid auditing.
Ask AI about this MCP
Supported MCP Clients
Gemini Waiting for input…
MongoDB Atlas Vector Search: 6 Tools for Data Management
These six tools let you manage the entire lifecycle of data in MongoDB Atlas—from listing collections to running complex vector similarity searches.
019d75d8create index
Creates a standard embedding Search Index bound to specific dimensions in your collection.
019d75d8delete
Deletes documents from a target collection, using parsed MongoDB filters to specify the data set.
019d75d8find
Retrieves standard MongoDB documents by resolving specific query filters against your collections.
019d75d8insert
Adds a new, generic document into the designated target collection within Atlas.
019d75d8list collections
Lists all accessible data collections that are explicitly managed and bounded inside your Atlas limits.
019d75d8search
Performs high-dimensional Vector similarity search using the advanced $vectorSearch operator.
Choose How to Get Started
Build a custom MCP for your own tools, or connect a ready-made integration from our catalog.
Build Your Own
Turn any API into an MCP. Import a spec, define Agent Skills, or deploy with MCPFusion.
- Import from OpenAPI, Swagger, or YAML specs
- Create Agent Skills with progressive disclosure
- Deploy to edge with MCPFusion framework
- Built in DLP, auth, and compliance on every call
- Real time usage dashboard and cost metering
- Publish to catalog or keep private
Make Your AI Do More
Start with MongoDB Atlas Vector Search, then connect any of our 4,700+ other servers whenever your AI needs more. One click, no limits.
- Use this MCP plus 4,700+ others, all in one place
- Add new capabilities to your AI anytime you want
- Every connection is secured and compliant automatically
- Track usage and costs across all your servers
- Works with Claude, ChatGPT, Cursor, and more
- New servers added to the catalog every week
What you can do with this MCP connector
You're getting full operational control over your MongoDB Atlas cluster right from your AI agent using this server. It lets your agent talk directly to your data, handling high-performance vector search and all kinds of standard document management. You don't need any manual scripting; your agent just runs the queries against the MCP standard.
When you use this setup, you can run advanced $vectorSearch queries using raw embedding vectors across your specified collections. This is how your AI client pulls back results that are semantically relevant to what it's looking for—it goes way beyond keyword matching. You'll execute sophisticated searches against the entire Atlas cluster.
For managing standard data, you use MQL filters for everything. Your agent can run find to pull specific MongoDB documents from any target collection by resolving exact query filters. If your system needs fresh data, it uses insert to drop a new, generic document into whatever collection you point it at within Atlas.
Conversely, if the data is stale or wrong, your agent runs delete, using parsed MongoDB filters to specify exactly which records need to go.
You're also in charge of setting up the infrastructure for these searches. To optimize performance, you can use create_index to build and configure a standard embedding Search Index. You define this index by specifying custom dimensions and mapping definitions that tell Atlas how to calculate similarity scores across your data.
Before starting, though, you need to know what's available. The agent runs list_collections to give you a complete list of every accessible data collection within the defined Atlas limits.
This system lets your agent run targeted MQL queries for rapid auditing by using find. This capability fetches specific metadata or structural chunks that bypass vector logic entirely, letting you check on raw data points fast. You're not just limited to searching vectors; you can query the underlying structure of your database too.
Basically, everything needed to manage your operational records—the finding, inserting, deleting, and indexing—is all housed here for your AI client to use. The agent handles connecting through the MCP standard and gives back structured data results instantly. You won't get a wall of text; you'll just get clean JSON output that tells you exactly what it found.
How MongoDB Atlas Vector Search MCP Works
- 1 Subscribe to the server and provide your MongoDB Atlas Data API URL and API Key.
- 2 Your AI client calls a tool (e.g.,
search) and passes the necessary query parameters, such as raw embedding vectors or MQL filters. - 3 The MCP Server executes the operation against your cluster and returns structured data results directly to your agent.
The bottom line is: you tell your AI client what to look for, and it runs the precise MongoDB Atlas command to get the answer.
Who Is MongoDB Atlas Vector Search MCP For?
ML Engineers who need to test vector relevance without writing boilerplate SDK code. Backend Developers managing operational data alongside vectors. Search Architects needing a unified way to audit indices and collection schemas.
Tests vector relevance and verifies embedding dimensions through natural conversation rather than manual script execution.
Manages operational documents (find/insert) and integrates the results with vector search in a single workflow from their terminal.
Audits complex search indices and monitors collection organization across multiple Atlas environments efficiently to ensure compliance.
What Changes When You Connect
- Seamlessly run vector searches: Use the
searchtool to execute sophisticated $vectorSearch queries, pulling back semantically relevant matches using raw embedding vectors. This eliminates manual SDK calls for basic retrieval. - Maintain data integrity: The
find,insert, anddeletetools let you manage standard MQL documents alongside your vector data. You treat the entire dataset—metadata included—as one system. - Build better search infrastructure: Use
create_indexto build Atlas Search indices. This gives you control over custom dimensions, optimizing how similarity calculations run. - Keep track of everything: The
list_collectionstool lets your agent audit your environment by listing every accessible collection name. You always know what data sources are available. - Verify data structure quickly: Use the
findor targeted MQL queries to fetch specific metadata chunks, bypassing vector logic for rapid structural verification and auditing.
Real-World Use Cases
Debugging a Knowledge Base Query
The ML Engineer needs to see if their embeddings are working correctly. They don't just run search; they first use the list_collections tool to confirm the correct namespace, then use find to pull metadata for 10 random IDs, and finally execute a targeted search using a known vector. This verifies both structure and relevance.
Updating User Records with New Vectors
The Backend Developer needs to update user profiles after training new embeddings. They first run find to pull the existing document IDs, then use the insert tool (or a conceptual upsert) to push the updated JSON record containing the fresh vector data.
Auditing Search Capabilities
The Search Architect suspects one collection is missing necessary indexing. They run list_collections to confirm existence, then use create_index to define a new Atlas Search index, and finally verify the change by attempting a complex search using the search tool.
Cleaning up Old Data
The data steward needs to delete records for users who left the company. They run find with an outdated status filter, and then use the delete tool on the resulting IDs list. This ensures both vector and operational data are purged.
The Tradeoffs
Assuming simple keyword search works.
Asking the agent to 'find all documents about cloud computing' without specifying a vector or structure, which leads to vague results from basic MQL queries.
→
You must use the search tool and provide a raw embedding vector. This forces the system to perform high-dimensional semantic matching instead of relying on simple keyword matches.
Ignoring schema changes when indexing.
Running create_index without first running list_collections to check the current structure, resulting in an index that fails because a required field doesn't exist or is misnamed.
→
Always run list_collections first. Then use the schema information to build your index correctly with create_index, ensuring all dimensions are properly mapped.
Mixing up data flow between tools.
Trying to update a document by running find and then manually passing the output JSON back into an unrelated tool, which breaks the transactional link.
→
Use find or list_collections to get the identifiers. Then use insert or delete with those identifiers to perform the specific write operation.
When It Fits, When It Doesn't
Use this server if your data retrieval requires knowing how similar two concepts are (vector search) AND you need to manage the underlying metadata and structure (MQL). Don't use it if you only need simple key-value lookups—a standard database connector is faster. Conversely, don't rely on find for semantic meaning; always prefer the search tool when dealing with embedded vectors. If your primary goal is just auditing what collections exist, start with list_collections. For any write operation (insert/delete), you must first confirm the target collection name using list_collections to prevent errors.
Independent Platform Disclaimer: Vinkius is an independent platform and is not affiliated with, endorsed by, sponsored by, verified by, or otherwise authorized by MongoDB Atlas Vector Search. All third-party trademarks, logos, and brand names are the property of their respective owners. Their use on this website is strictly for informational purposes to identify service compatibility and interoperability.
VINKIUS INFRASTRUCTURE
Cloud Hosted
Managed infra
V8 Isolated
Sandboxed per request
Zero-Trust Proxy
No stored credentials
DLP Enforced
Policy on every call
GDPR Compliant
EU data residency
Token Compression
~60% cost reduction
Works with Claude, ChatGPT, Cursor, and more
The Model Context Protocol standardizes how applications expose capabilities to LLMs. Instead of operating in isolation, your AI gains direct access to external platforms, live data, and real-world actions through secure, standardized connections.
This server provides 6 capabilities that interface natively with Claude, ChatGPT, Cursor, and any MCP client. No middleware. No custom integration required.
Available Capabilities
Manually verifying data structure and indices is a huge time sink.
Today, checking if your search index needs updating means jumping between the Atlas GUI, running multiple MQL queries just to check the schema boundaries, and cross-referencing documentation. It's tedious clicking—you spend more time verifying structure than improving it.
With this MCP server, you simply ask: 'What collections exist?' or 'Show me the current search index definitions.' The agent runs `list_collections` or uses other tools to provide the schema boundaries directly in your chat. You get immediate structural confirmation.
MongoDB Atlas Vector Search MCP Server gives you full data control.
Before, updating a record meant running multiple steps: first finding the old ID via `find`, then manually pulling the new vector into a separate editor, and finally executing a write command. It was slow and prone to copy/paste errors.
Now, you just ask the agent to update the document using its ID. The agent handles the logic—it executes the necessary sequence of reads, processes the data, and runs `insert` or `delete` in one seamless flow. Data changes are atomic.
Common Questions About MongoDB Atlas Vector Search MCP
How do I check which collections are available using MongoDB Atlas Vector Search MCP Server? +
Run the list_collections tool. This immediately tells you all accessible data collections within your defined Atlas limits, so you know exactly what namespaces exist.
What is the difference between `find` and `search` with MongoDB Atlas Vector Search MCP Server? +
Find uses standard MQL filters to look for documents based on explicit field values (e.g., 'status: active'). Search runs high-dimensional vector similarity, finding semantically related matches even if the keywords don't match exactly.
Can I create an index using only the name of the collection? +
No. You must use create_index and specify both the target collection name AND the exact dimensions for the binding index to work correctly.
If I update a document, which tool should I use: `find` or `insert`? +
Use insert. While you might first run find to get the ID, insert is the action that writes the updated data (the new JSON record) into the collection.
What credentials do I need to successfully run the `search` tool? +
You must provide your MongoDB Atlas Data API URL and an active API Key. These credentials grant your agent access to the cluster, allowing it to execute vector similarity searches and manage data operations.
How does using the `create_index` tool improve my search performance? +
The create_index tool optimizes your cluster's internal infrastructure. By defining custom dimensions and mapping, you pre-calculate the necessary structures for similarity checks, making subsequent vector searches much faster.
When using the `delete` tool, what determines which documents are removed? +
The deletion is governed by parsed MongoDB filters. You specify exactly which records you want gone—the system doesn't guess; it removes documents that match your literal MQL query filters.
Does `list_collections` provide schema information, or just collection names? +
list_collections retrieves the name of every managed data set in Atlas. More importantly, it also gives you access to the schema boundaries, letting you audit the structure and organization across your database.
Can I manage both vector search and standard data in the same conversation? +
Yes. MongoDB Atlas Vector Search is unified. You can use the search tool for similarity and the find or insert tools for standard operational data management using MQL, allowing you to bridge both worlds natively.
How do I create a new vector search index through the agent? +
Use the create_index tool by providing the database, collection, and required dimensions (matching your embedding model). Your agent will provision the index infrastructure on Atlas to enable high-speed vector retrieval.
Can my agent find specific documents using standard MongoDB query filters? +
Absolutely. Use the find tool with a JSON string representing your MQL filter (e.g. {"status":"active"}). Your agent will execute the Data API request and return the matching documents and their scalar properties securely.
Use it with your favorite AI tools
Connect this server to Cursor, Claude, VS Code, and more.
More in this category
Mercado Livre Orders
Search sales, manage shipments, track packages, and handle cancellations on Mercado Livre.
ScreenshotAPI
Capture full-page website screenshots programmatically with custom viewport sizes, delays, and rendering options via a simple API.
iFLYTEK Open Platform / 讯飞开放平台
China's leading voice and NLP platform — convert speech to text, synthesize voice, and analyze text via AI.
You might also like
Stannp
Automate direct mail and physical letters via Stannp — manage letters, postcards, and campaigns directly from any AI agent.
RMSE & MAE Calculator
Compute exact Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) for regression models. Stop hallucinating model validation metrics.
Kelkoo
Search product offers, compare prices, and explore retail catalogs via Kelkoo.