A MongoDB Atlas MCP for automation that reaches past the cluster
Connect one organization API key pair and cluster lifecycle, database users, network access, cloud backups, alerts, and monitoring measurements become workflow steps. An AI agent can pause development clusters on a schedule, grant database access that expires on its own, and read a cluster's metrics before anyone gets paged, with the results landing next to your tasks and databases instead of in a terminal.
What the MongoDB Atlas MCP connector does
The MongoDB Atlas connector for Mattering Automations is 44 nodes across the Atlas Administration API v2: organizations and projects, dedicated cluster lifecycle (create, update, pause, resume, delete), Flex clusters (the current low-cost generation that replaced serverless instances), database user management, the project IP access list, cloud backup snapshots and restore jobs, alerts with acknowledge, read-only alert configurations, project and organization activity events, process and disk measurements, teams, and organization API key listing. One organization API key pair authenticates all of it over HTTP Digest.
Every node speaks MCP, the open protocol for connecting AI assistants to tools, so Claude, ChatGPT, and any MCP-capable assistant can use them, alongside Matter, the AI assistant built into the platform. An agent can list a project's clusters, check their metrics against a threshold, pause the ones nobody is using, and write what it did into a Mattering database, without anyone opening the Atlas console.
MongoDB maintains its own official MCP server, and it is good at what it is for: giving an assistant a direct line to MongoDB. This connector is for the other half of the job, where Atlas is one step among many. On the Mattering canvas the same nodes sit next to every other connector and the platform itself, so an open alert can become a triage task, a restore can wait behind a human approval step, and a nightly metrics sweep can land its findings in a database next to the rest of your work. Use whichever fits, or both.
The safety modeling is stated plainly. Of the 44 nodes, 28 read and 16 act, and five are flagged destructive: project delete, cluster delete, Flex cluster delete, database user delete, and restore jobs, because an automated or point-in-time restore replaces all data on the target cluster. Some surfaces are deliberately not shipped at all: API key creation stays an operator act (the connector only lists keys), alert configuration rules are read-only because a mis-authored matcher can silently disable paging, serverless instances are omitted because MongoDB deprecated them in favor of Flex, and your database documents are out of scope entirely. This is the control plane, held at the surface an automation can hold safely.
One connection, three surfaces
- Visual automations. Drag MongoDB Atlas nodes onto the canvas and wire them to triggers, conditions, and other apps.
- AI agents, yours or ours. These nodes speak MCP, the open protocol, so Claude, ChatGPT, and any MCP-capable assistant can use them, and Matter, the agent built into the platform, uses them natively.
- The whole platform. Combine with native tasks, projects, and other connectors, no middleware in between.
Every MongoDB Atlas node, ready to drop on the canvas
44 nodes, generated from the connector itself so this list always matches what ships. Expand a category to browse.
Read and query nodes 28
MongoDB Atlas: Get Access List Entry
Get one access list entry by its value - the IP address (192.0.2.10), CIDR block (192.0.2.0/24), or AWS security group ID exactly as it appears in the list. The runner URL-encodes the value (slashes in CIDR blocks encode cleanly).
MongoDB Atlas: Get Alert
Get one alert's details: status, event type, current metric value, the cluster or host it fired on, and acknowledgement state.
MongoDB Atlas: Get Alert Configuration
Get one alert configuration's full rule: event type, enabled flag, matchers, metric threshold, and notification channels.
MongoDB Atlas: Get Backup Snapshot
Get one cloud-backup snapshot's details: status, creation and expiry times, MongoDB version at snapshot time, and storage size.
MongoDB Atlas: Get Cluster
Get one cluster's full configuration by name: state, version, paused flag, backup settings, connection strings, and replication specs. Poll stateName (IDLE / CREATING / UPDATING / REPAIRING / DELETING) to wait for a change to land.
MongoDB Atlas: Get Database User
Get one database user by auth database and username: roles, scopes, and external-auth type flags. The password is never returned.
MongoDB Atlas: Get Disk Measurements
Get timeseries disk measurements for one partition on a process host: free and used space, IOPS, and latency, at a chosen granularity. Same window rules as process measurements: set EITHER period OR start+end - never both.
MongoDB Atlas: Get Flex Cluster
Get one Flex cluster's details by name: state, MongoDB version, backing provider and region, backup settings, and connection strings.
MongoDB Atlas: Get Organization
Get one organization's details by ID: name, deletion state, and default-alert-settings flag.
MongoDB Atlas: Get Process Measurements
Get timeseries measurements for one MongoDB process: connections, opcounters, CPU, memory, cache activity, and more, at a chosen granularity. Each measurement returns dataPoints of timestamp/value pairs. Set EITHER period (rolling lookback) OR start+end (fixed window) - never both. The core observability read for threshold checks and reporting automations.
MongoDB Atlas: Get Project
Get one project's details by ID: name, owning organization, cluster count, and creation date.
MongoDB Atlas: Get Team
Get one team's details by ID (id and name; member listing is a separate users endpoint outside this connector's scope).
MongoDB Atlas: List API Keys
List an organization's programmatic API keys with their descriptions, public keys, and role assignments - the credential audit read. Atlas partially redacts each privateKey in this response; full private keys are shown exactly once at creation in the UI. Key creation is deliberately not available through this connector.
MongoDB Atlas: List Access List Entries
List the project's IP access list: the IP addresses, CIDR blocks, and AWS security groups allowed to connect to the project's clusters, with comments and any auto-expiry dates.
MongoDB Atlas: List Alert Configurations
List a project's alert configurations (the rules that fire alerts) with event types, matchers, thresholds, and notification channels. Read-only surface by scope - authoring alert rules stays in the Atlas UI.
MongoDB Atlas: List Alerts
List a project's alerts with status, event type, the triggering metric and value, and acknowledgement state. Filter by status OPEN, TRACKING, or CLOSED. The alert ID from here feeds Get Alert and Acknowledge Alert.
MongoDB Atlas: List Backup Snapshots
List a cluster's replica-set cloud-backup snapshots with creation and expiry times, status, type (scheduled or onDemand), and storage size. The snapshot ID from here feeds Get Backup Snapshot and Create Restore Job. Covers replica-set clusters; sharded-cluster snapshots live on a separate endpoint that is out of scope.
MongoDB Atlas: List Clusters
List the dedicated clusters in a project with their names, states, MongoDB versions, paused flags, connection strings, and replication specs. The cluster NAME from here addresses every other cluster node - Atlas has no cluster-id path.
MongoDB Atlas: List Database Users
List the database users in a project with their auth databases, roles, and cluster scopes. Passwords are never returned. The username from here feeds the get / update / delete user nodes.
MongoDB Atlas: List Disks
List the disk partitions on one MongoDB process host (typically a data partition). The partition name from here feeds Get Disk Measurements.
MongoDB Atlas: List Flex Clusters
List the Flex clusters in a project (the low-cost tier that replaced serverless instances) with names, states, provider settings, and connection strings. Dedicated clusters live on List Clusters instead.
MongoDB Atlas: List Organization Events
List an organization's activity events (billing changes, membership changes, project creation) filtered by date range and event types. Append-only audit entries at org scope.
MongoDB Atlas: List Organizations
List the organizations the API key can access with their ids and names. The cheapest authenticated read on the whole API and the credential health check to run first in any Atlas automation - the organization ID from here feeds the org-scoped nodes (events, teams, API keys).
MongoDB Atlas: List Processes
List the MongoDB processes (mongod and mongos nodes) running in a project, each identified as hostname:port with its role (REPLICA_PRIMARY, REPLICA_SECONDARY, SHARD_MONGOS...). The process ID from here feeds the measurement and disk nodes.
MongoDB Atlas: List Project Events
List a project's activity events (cluster changes, user actions, alert firings) filtered by date range, event types, or cluster names. Events are append-only audit entries - the compliance and change-feed read.
MongoDB Atlas: List Projects
List the projects (Atlas 'groups') the API key can access across all its organizations, with ids, names, org ids, and cluster counts. The project ID from here feeds every project-scoped node: clusters, database users, access lists, backups, alerts, events, and measurements.
MongoDB Atlas: List Restore Jobs
List a cluster's backup restore jobs with delivery type, source snapshot, target cluster, and completion state. The monitoring read for restore automations - poll until finishedAt is set or failed flips true.
MongoDB Atlas: List Teams
List an organization's teams with their ids and names. Teams group Atlas users for project role assignment; the team ID from here feeds Get Team.
Action nodes 16
MongoDB Atlas: Acknowledge Alert
Acknowledge an alert until a chosen time (notifications pause until then) or un-acknowledge it by sending unacknowledgeAlert true. Fully reversible - the alert record itself is untouched.
MongoDB Atlas: Add Access List Entries
Add one or more entries to the project's IP access list. The request body is an ARRAY of entry objects; each entry carries exactly one of ipAddress, cidrBlock, or awsSecurityGroup, plus an optional comment and an optional deleteAfterDate for temporary access (max 7 days out). Reversible - Delete Access List Entry removes an entry.
MongoDB Atlas: Create Cluster
Create a dedicated cluster in a project. Provisions billable cloud infrastructure that keeps running until deleted - pair test creates with Delete Cluster. Creation is asynchronous: the response returns stateName CREATING and the cluster becomes usable minutes later (poll Get Cluster).
MongoDB Atlas: Create Database User
Create a database user in a project. Users apply to every cluster in the project unless narrowed with scopes. Reversible - Delete Database User removes it. Supports temporary users via deleteAfterDate (auto-removed at that time, max 7 days out).
MongoDB Atlas: Create Flex Cluster
Create a Flex cluster in a project - the low-cost tier for development and small workloads. Provisions a billable resource that runs until deleted; pair test creates with Delete Flex Cluster. Creation is asynchronous (stateName CREATING, then IDLE).
MongoDB Atlas: Create Project
Create a new project inside an organization. Projects are the container for clusters, database users, and network access rules. Reversible - Delete Project removes an empty project cleanly.
MongoDB Atlas: Create Restore Job
DESTRUCTIVE: restore a snapshot. With deliveryType automated or pointInTime this REPLACES ALL DATA on the target cluster with the snapshot's data; deliveryType download instead produces a download URL and touches nothing. cluster_name is the SOURCE cluster whose backup is restored; the target fields say where the data lands.
MongoDB Atlas: Delete Access List Entry
Remove one entry from the project's IP access list by its value (IP address, CIDR block, or AWS security group ID). Clients connecting from that range are refused until it is re-added; recreating the entry is a single Add call with the same value.
MongoDB Atlas: Delete Cluster
DESTRUCTIVE: permanently delete a cluster and all data on it. Set retainBackups true to keep existing cloud-backup snapshots after the cluster is gone (restorable into a new cluster); otherwise snapshots are deleted too. Refused while terminationProtectionEnabled is true. Deletion is asynchronous (202 Accepted).
MongoDB Atlas: Delete Database User
DESTRUCTIVE: delete a database user. Applications authenticating as this user lose access immediately. The user record is recreatable but its password is not recoverable, so recreation means issuing a new credential.
MongoDB Atlas: Delete Flex Cluster
DESTRUCTIVE: permanently delete a Flex cluster and all data on it. Refused while terminationProtectionEnabled is true. Deletion is asynchronous (202 Accepted).
MongoDB Atlas: Delete Project
DESTRUCTIVE: permanently delete a project. The project must be empty first (no clusters); Atlas refuses otherwise. The project record, its access lists, database users, and alert configurations are gone irreversibly.
MongoDB Atlas: Pause Cluster
Pause a dedicated cluster (PATCH with paused true - the v2 API has no separate pause endpoint). A paused cluster stops compute billing but keeps storage; connections are refused while paused. Atlas auto-resumes a paused cluster after 30 days unless it is paused again. Resume Cluster reverses this. Only M10+ dedicated clusters can be paused, and a freshly resumed cluster must run ~60 minutes before it can pause again.
MongoDB Atlas: Resume Cluster
Resume a paused dedicated cluster (PATCH with paused false). Compute billing restarts; the cluster accepts connections again once stateName returns to IDLE.
MongoDB Atlas: Update Cluster
Update a cluster's configuration: scale instance sizes via replicationSpecs, toggle backups or termination protection, change the MongoDB major version, or set tags. PATCH semantics - only the fields sent change. Use the dedicated Pause Cluster / Resume Cluster nodes for the paused flag. Changes apply asynchronously (poll Get Cluster).
MongoDB Atlas: Update Database User
Update a database user's password, roles, scopes, or description. Only the fields sent change. The username and auth database cannot be changed - delete and recreate for that.
Connecting MongoDB Atlas takes one credential
Atlas authenticates Administration API requests with an organization API key pair sent over HTTP Digest: the public key acts as the username and the private key as the password. Keys are created in the Atlas console on the organization's Applications page, found under Identity & Access in the sidebar, and each key carries the organization roles you assign at creation and can be restricted to an IP access list. MongoDB's documentation describes API keys as a legacy authentication method and recommends service accounts for new integrations, but keys remain fully supported with no announced sunset, and the static two-field pair is the credential this connector uses.
Sign in at cloud.mongodb.com, select the organization, click Applications in the sidebar under Identity & Access, and click Add new API Key. Give it a description you will recognize and assign the organization roles the key needs: a read-only role covers the 28 query nodes, while management workflows need roles that can modify the projects and clusters they touch.
Copy the private key when it is shown. Atlas displays it once, at creation; the public key stays visible and is not a secret.
In Mattering Automations, open Connectors, click Add Connector, choose MongoDB Atlas, and create a credential with the public key and the private key.
Click Test Connection. Mattering calls the Atlas organizations endpoint, the cheapest authenticated read on the API, driving the full Digest handshake with the versioned Accept header Atlas requires. Success proves the key pair; a 403 with a correct key usually means the key carries an IP access list that does not include the address the call came from.
All 44 MongoDB Atlas nodes appear in the canvas palette. Run MongoDB Atlas: List Organizations first for the organization ID, then MongoDB Atlas: List Projects for the project ID that every project-scoped node needs.
Common uses for the MongoDB Atlas MCP
Development clusters that pause themselves
MongoDB Atlas: Pause Cluster stops a dedicated cluster's compute billing while keeping its storage, and MongoDB Atlas: Resume Cluster brings it back, so a scheduled workflow can pause development and staging clusters every evening and resume them each morning. The mechanics worth knowing are on the nodes: only M10 and larger dedicated clusters pause, Atlas auto-resumes a paused cluster after 30 days unless it is paused again, and a freshly resumed cluster must run about 60 minutes before it can pause again. Both nodes are fully reversible, which is what makes them safe to hand to a schedule.
Temporary access that cleans up after itself
MongoDB Atlas: Add Access List Entries opens the project's IP access list for an address or CIDR block with an optional deleteAfterDate, up to 7 days out, so the network grant removes itself. MongoDB Atlas: Create Database User supports the same deleteAfterDate for temporary users, and scopes can narrow a user to specific clusters. A workflow can grant a contractor or a CI runner exactly the access it needs and let Atlas do the revoking, with MongoDB Atlas: List Access List Entries and MongoDB Atlas: List Database Users as the audit reads.
An agent that checks the metrics before anyone gets paged
MongoDB Atlas: List Processes identifies every mongod and mongos process in a project, and MongoDB Atlas: Get Process Measurements returns timeseries data for connections, opcounters, CPU, memory, and cache activity at a granularity you choose. MongoDB Atlas: List Disks and MongoDB Atlas: Get Disk Measurements cover free space, IOPS, and latency per partition. A scheduled sweep can compare the numbers against thresholds, write the results into a Mattering database, and open a task only when something crosses a line. One rule the API enforces: set either a rolling period or a fixed start and end window, never both.
Alert triage that lands in your task list
MongoDB Atlas: List Alerts filtered to status OPEN shows what is firing, MongoDB Atlas: Get Alert pulls one alert's full detail, and MongoDB Atlas: Acknowledge Alert pauses its notifications until a time you choose, with a comment, and can un-acknowledge later, so the acknowledgement is fully reversible. A workflow can turn each new open alert into a Mattering Productivity task carrying the cluster name and metric value, acknowledge the alert so the pager quiets down, and leave the task as the record of who is on it. MongoDB Atlas: List Alert Configurations shows the rules behind the alerts, read-only on purpose.
Ephemeral environments provisioned end to end
MongoDB Atlas: Create Project makes the container, MongoDB Atlas: Create Flex Cluster provisions a low-cost cluster inside it, MongoDB Atlas: Create Database User adds credentials, and MongoDB Atlas: Add Access List Entries opens the network path. Cluster creation is asynchronous, so the workflow polls MongoDB Atlas: Get Flex Cluster until stateName reads IDLE and then hands the connection string onward. The teardown is the same shape in reverse: MongoDB Atlas: Delete Flex Cluster, then MongoDB Atlas: Delete Project once the project is empty, both flagged destructive so workflows gate them.
Backup awareness and gated restores
MongoDB Atlas: List Backup Snapshots shows a cluster's cloud-backup snapshots with their creation and expiry times, so a scheduled check can open a task when the newest snapshot is older than it should be. MongoDB Atlas: Create Restore Job is flagged destructive for a reason: with deliveryType automated or pointInTime it replaces all data on the target cluster, which is the point of a restore and exactly why it belongs behind an approval step. The download delivery type instead produces a download URL and touches nothing, which makes it the safe way to pull snapshot data out. MongoDB Atlas: List Restore Jobs is the poll that watches a restore run to completion.
Example: A pull request opens: provision a Flex cluster, a scoped user, and access that expires on its own
Built on the Automations canvas in a few minutes, or described in plain language to Matter and built for you.
Trigger: an inbound webhook from your CI pipeline fires when a pull request opens, carrying the branch name and the test runner's IP address.
MongoDB Atlas: Create Project creates a project named for the branch, the container the rest of the run fills.
MongoDB Atlas: Create Flex Cluster provisions a low-cost cluster in it. Creation is asynchronous: the response comes back with stateName CREATING.
MongoDB Atlas: Get Flex Cluster polls until stateName reads IDLE and returns the connection strings.
MongoDB Atlas: Create Database User creates a user for the branch, and MongoDB Atlas: Add Access List Entries opens the runner's IP with a deleteAfterDate, so the network grant revokes itself within the week.
The workflow posts the connection string and cluster details to a task in Mattering Productivity; when the pull request closes, a second automation runs MongoDB Atlas: Delete Flex Cluster and then MongoDB Atlas: Delete Project to tear the whole environment down.
MongoDB Atlas MCP questions, answered
What MongoDB Atlas operations are supported?
44 nodes on the Atlas Administration API v2: 28 query nodes and 16 action nodes covering organizations and projects, dedicated cluster lifecycle (create, update, pause, resume, delete), Flex clusters, database user management, the project IP access list, cloud backup snapshots and restore jobs, alerts with acknowledge, read-only alert configurations, project and organization activity events, process and disk measurements, teams, and organization API key listing. Deliberately not shipped: API key creation, alert configuration writes, serverless instances (deprecated in favor of Flex), sharded-cluster snapshot variants, and anything that touches the documents inside your databases.
Does this work with Claude, ChatGPT, or other AI assistants?
Yes. Mattering exposes these nodes over MCP, an open protocol for connecting AI assistants to tools, so any MCP-capable assistant, including Claude and ChatGPT, can connect to your workspace and use them. Matter, the assistant built into the platform, uses the same nodes natively.
MongoDB publishes an official MCP server. Why use this connector?
MongoDB does maintain an official MCP server, and if all you want is an assistant talking to MongoDB alone, it is a genuine option. This connector earns its place when Atlas is one step in a larger workflow: the same canvas holds your other connectors and the Mattering platform itself, so an open alert can become a triage task, a restore can wait behind a human approval step, and a nightly metrics sweep can write into a database next to the rest of your work. Credentials live in one place with your other connector secrets, and the same nodes serve visual automations and AI agents alike. Use whichever fits, or both.
Can an automation read or write documents in my collections?
No, and that boundary is deliberate. This connector covers the Atlas Administration API, the control plane: clusters, database users, network access, backups, alerts, events, and metrics. It never connects to the databases themselves, so reading and writing collection data happens through MongoDB drivers and tools outside this connector's scope. The practical upside is that the credential you hand an automation manages infrastructure without also being a path to the data in it.
How does authentication work?
With an organization API key pair over HTTP Digest: the public key is the username, the private key is the password, and Mattering's runner drives the challenge handshake on every request along with the versioned Accept header the v2 API requires. MongoDB's documentation describes API keys as a legacy authentication method and recommends service accounts for new integrations, but keys remain fully supported with no announced sunset, and the static two-field pair is what this connector uses. Keys carry the organization roles you assign at creation and can be restricted to an IP access list, in which case calls from other addresses get a 403 from Atlas.
Can an AI agent safely manage Atlas clusters?
That is one of the connector's designed-for jobs, and the guardrails are real rather than implied. The five destructive nodes (project, cluster, Flex cluster, and database user deletes, plus restore jobs) are flagged so workflows gate them behind conditions or approval steps. Atlas's own protections are honored too: a cluster with termination protection enabled refuses deletion, and MongoDB Atlas: Delete Cluster can set retainBackups so snapshots survive the cluster. The reversible surface is generous, which is where an agent should live most of the time: pause and resume undo each other, acknowledged alerts can be un-acknowledged, and access list entries are recreatable with a single call.
What happened to serverless instances?
MongoDB deprecated them and has since retired the tier: Atlas migrated existing serverless instances to Free, Flex, or Dedicated clusters according to usage. The vendor's published API specification marks the whole serverless endpoint family deprecated, with most of the endpoints already sunset as of January 22, 2026 and the remaining serverless instance reads carrying a sunset date of January 18, 2027. This connector deliberately ships the Flex cluster nodes (list, get, create, delete) and does not include the serverless endpoints at all, so automations built on it are on the surface MongoDB is keeping.
What are MongoDB Atlas's API rate limits?
Atlas documents rate limits per endpoint set rather than as one global number, and requests over a limit return HTTP 429 with rate-limit headers. The nodes report the 429 on their output rather than silently retrying, so schedule wide sweeps apart and lean on pagination: list endpoints accept up to 500 items per page, which usually turns a burst of Get calls into one List call.
Put MongoDB Atlas to work inside your workspace
Connect once, then use every node in visual automations and AI agent workflows across the Mattering platform.
All product names, logos, and brands are property of their respective owners and are used for identification purposes only. Use of these names, logos, and brands does not imply endorsement or affiliation. MongoDB Atlas is a product of its respective owner; Mattering is an independent platform.