MongoDB Atlas

MongoDB Atlas Expert

Generated from docs/mongodb-atlas-expert-context.md in 10gen/mdb-tam. Use it as a MongoDB Atlas platform reference when planning Atlas architecture, automating Atlas administration, connecting applications, designing Atlas-backed schemas and queries, or reviewing Atlas operational posture. Start from the context below, then defer to the linked official MongoDB Atlas docs, MongoDB Manual, and driver docs as the source of truth for exact endpoint, command, operator, and version details. For a deep sub-area, match the task to the Sub-skill routing table below and read the listed references/…md file before answering.

Sub-skill routing table

This skill consolidates 27 Atlas sub-skills as on-demand references — match the task to the table and Read the listed references/…md file before answering deep questions. Do not rely on this table alone for depth.

Source scope

Atlas platform and architecture

• Atlas docs home: deployment types, regions, access control, connection, alerts, optimization entry points
  https://www.mongodb.com/docs/atlas/
• Atlas Architecture Center: Atlas well-architected guidance across operational efficiency, security, reliability, performance, and cost optimization
  https://www.mongodb.com/docs/atlas/architecture/current/

Atlas administration and automation

• Atlas Admin API v2 reference: canonical Atlas administration endpoint inventory and auth model
  https://www.mongodb.com/docs/api/doc/atlas-admin-api-v2/
• Configure Atlas API access: service-account/API-key setup, IP access list behavior, and the REST/data-plane boundary
  https://www.mongodb.com/docs/atlas/configure-api-access/
• Atlas CLI docs: command-line Atlas management surface and setup flow
  https://www.mongodb.com/docs/atlas/cli/current/
• Atlas Terraform provider guide: infrastructure-as-code surface for Atlas provisioning and lifecycle management
  https://www.mongodb.com/docs/atlas/terraform/
• Atlas Kubernetes Operator: Kubernetes control-plane integration and lifecycle caveats
  https://www.mongodb.com/docs/atlas/operator/stable/

Connection, security, and access control

• Connect to a database deployment: connection prerequisites, private networking choices, firewall requirements, and data-plane connection flow
  https://www.mongodb.com/docs/atlas/connect-to-database-deployment/
• Atlas IP access list: project-scoped client network allow-list behavior, limits, temporary entries, and CLI entry points
  https://www.mongodb.com/docs/atlas/security/ip-access-list/
• Atlas database users: Atlas-vs-database-user boundary, auth mechanisms, role model, and operational limits
  https://www.mongodb.com/docs/atlas/security-add-mongodb-users/

Data access and MongoDB application design

• MongoDB drivers: official application client surfaces by language
  https://www.mongodb.com/docs/drivers/
• MongoDB data modeling: access-pattern-first schema design, embedding vs referencing, and flexible-schema guidance
  https://www.mongodb.com/docs/manual/data-modeling/
• MongoDB indexes: index types, write/read tradeoffs, Atlas UI/CLI index management, and Performance Advisor entry points
  https://www.mongodb.com/docs/manual/indexes/
• MongoDB aggregation: preferred aggregation-pipeline model and stage-based data processing
  https://www.mongodb.com/docs/manual/aggregation/

Atlas search and AI surfaces

• Atlas Search: full-text search, analyzers, mappings, $search, $searchMeta, pagination, faceting, and autocomplete
  https://www.mongodb.com/docs/atlas/atlas-search/
• Atlas Vector Search: ANN/ENN vector search, hybrid search, RAG, automated embedding, and version availability
  https://www.mongodb.com/docs/atlas/atlas-vector-search/

Streaming, encryption, and newer surfaces

• Atlas Stream Processing: continuous stream processing using aggregation-pipeline syntax over Atlas and Kafka sources
  https://www.mongodb.com/docs/atlas/atlas-stream-processing/
• Queryable Encryption: query encrypted fields without exposing plaintext; equality/range GA, prefix/suffix/substring in preview on 8.2+
  https://www.mongodb.com/docs/manual/core/queryable-encryption/
• Atlas AI Integrations: consolidated RAG, agent memory, and embedding pattern documentation
  https://www.mongodb.com/docs/atlas/ai-integrations/

Operations, backup, observability, and limits

• Atlas alerts: alert conditions, lifecycle, acknowledgement, and org/project settings
  https://www.mongodb.com/docs/atlas/alerts/
• Atlas cluster metrics: health metrics, real-time metrics, search metrics, and operator-facing signals
  https://www.mongodb.com/docs/atlas/monitor-cluster-metrics/
• Atlas Cloud Backup overview: backup enablement, redundancy, compliance, restore-role requirements, and topology caveats
  https://www.mongodb.com/docs/atlas/backup/cloud-backup/overview/
• Atlas limits: component, connection, and topology limits that shape designs and operating posture
  https://www.mongodb.com/docs/atlas/reference/atlas-limits/
• Atlas Performance Advisor: slow-query analysis, index suggestions, and read-vs-write tradeoff reminders
  https://www.mongodb.com/docs/atlas/performance-advisor/

Atlas quick rules

1. Treat Atlas as the control plane and MongoDB drivers / mongosh as the main data-plane application surfaces.
2. Prefer service accounts over legacy API keys for new Atlas Admin API automation.
3. Remember the Atlas Admin API does not read or write your cluster data; it manages Atlas resources and access configuration.
4. Design schemas around access patterns; data accessed together should generally live together.
5. Prefer embedding when it lets common reads complete as a single-document fetch; use transactions only when requirements truly cross document boundaries.
6. Add indexes for repeated query patterns, but account for the write cost of every index.
7. Use Performance Advisor, cluster metrics, alerts, and limits together; Atlas performance work is not just about query syntax.
8. Treat IP access lists, private networking, database users, and Atlas users as distinct security controls with different scopes.
9. Use Atlas Search for full-text/relevance workloads and Atlas Vector Search for semantic similarity/RAG workloads; they are related but not the same feature.
10. Check deployment tier, MongoDB version, and Atlas limits before giving prescriptive advice.
11. Know the Flex / Free / Dedicated tier model; Serverless and M2/M5 are deprecated.
12. Consider Atlas Stream Processing for event-driven and real-time ETL workloads before building external pipelines.
13. For hybrid retrieval, use $rankFusion (MongoDB 8.1+) to merge full-text and vector results in a single query.
14. Use Queryable Encryption for sensitive fields that still need to be queried; equality/range queries are GA, prefix/suffix/substring are in preview on 8.2+.

What Atlas expertise should mean

An Atlas-focused assistant should be able to reason across:

• control plane design: organizations, projects, deployments, users, network boundaries, backups, monitoring, limits, and automation
• application access: drivers, connection strings, connection pooling, query/index fit, aggregation, search, vector search
• operations: alerts, metrics, performance advisor, capacity, topology, and backup/disaster-recovery posture
• delivery models: UI, Admin API, CLI, Terraform, and Atlas Kubernetes Operator
• streaming and event-driven: Atlas Stream Processing for real-time pipelines and time-series ingestion
• AI and agentic patterns: automated embedding, vector search, hybrid search, RAG, agent memory, and framework integrations

Atlas operating model

Shared responsibility

• Atlas Architecture Center explicitly frames Atlas through a shared responsibility model: MongoDB operates the underlying platform, while customers own their configuration, access control, and data policies.
• Atlas architectural guidance is organized around five pillars: operational efficiency, security, reliability, performance, and cost optimization.

Core control-plane hierarchy

• Atlas operations are anchored around organizations, projects, and deployments/clusters.
• Many role assignments and operational actions are scoped differently at the organization vs project layer, so advice should always name the required scope.

Deployment and connectivity model

• Atlas lets teams choose deployment tier (Free, Flex, or Dedicated), cloud provider, and region based on application latency, cost, and security requirements. Serverless instances and M2/M5 clusters are deprecated; Flex is their replacement.
• Application environments must satisfy both network access and database-user authentication to reach an Atlas deployment.
• Private connectivity choices include VPC/VNet peering and private endpoints, while public connectivity requires IP allow-listing.

Atlas interaction methods inventory

This is the high-value inventory of how you can work with Atlas. For exact subcommands or endpoints, follow the linked source sections.

Atlas Admin API standards

• The Atlas Admin API is a REST-style control-plane API for Atlas resource administration.
• The preferred authentication method is OAuth2 service-account access tokens; HTTP Digest API keys are a legacy option.
• Service-account tokens are obtained from POST https://cloud.mongodb.com/api/oauth/token and are reusable for about one hour.
• If your organization requires an API IP access list, token generation can happen from any IP, but API calls that use the token must come from an allowed IP.
• The Admin API does not expose application data access; reading and writing database documents still happens through cluster auth plus a driver or other data-plane client.
• As of March 2026, the Admin API enforces standardized rate limiting using a token-bucket algorithm. Automation code must handle 429 Too Many Requests responses and respect Retry-After headers.

Admin API categories to expect

For exact endpoints, use the API reference. In practice, Atlas Admin API work usually falls into these buckets:

• organizations and projects
• deployments / clusters / topology
• database users and access
• network access and IP lists
• alerts and monitoring metadata
• backup and restore administration
• inventory, IP address, and project metadata

Atlas CLI standards

• Atlas CLI is the terminal-native Atlas control-plane surface.
• atlas setup is the fast-start command that signs up or authenticates, creates a free database, loads sample data, adds the current IP to the access list, creates a database user, and connects with mongosh.
• Atlas CLI is appropriate for operator workflows and scripts, but when a task needs stable, reviewable automation, the Admin API or Terraform may be a better fit.
• Atlas docs explicitly surface CLI entry points for actions like access list inspection and other Atlas resource operations.

Terraform and Kubernetes Operator standards

Terraform

• Use Terraform when you want declarative Atlas infrastructure management.
• The Atlas Terraform provider is the main Atlas IaC surface for provisioning/managing clusters and related Atlas resources from code.
• MongoDB’s guide expects service-account authentication for provider configuration.
• Provider 2.0 (2025) introduced semantic versioning with no-breaking-change guarantees in minor/patch releases. Projects on 1.x must follow the 2.0.0 Upgrade Guide before upgrading. See the refresh section below for details.

Atlas Kubernetes Operator

• Use Atlas Kubernetes Operator when Atlas resources need to be managed from a Kubernetes control plane.
• AKO manages Atlas state from custom resources like AtlasProject, AtlasDeployment, and AtlasDatabaseUser.
• AKO 2.0 changed deletion behavior: deleting a Kubernetes custom resource no longer deletes the Atlas resource by default.
• The docs explicitly warn to define desired config values explicitly to avoid inheriting Atlas defaults that can cause reconciliation loops.

Atlas security and access-control standards

Network boundaries

• Atlas only allows cluster client connections from the project’s IP access list unless you use private networking.
• IP access lists are project-wide, not per cluster.
• Atlas supports temporary IP access list entries with configurable expiration.
• For application connectivity, Atlas docs call out three main patterns: public IP allow-listing, VPC/VNet peering, and private endpoints.

Identity boundaries

• Atlas users are not the same as database users.
• Atlas users access the Atlas control plane; database users access MongoDB data-plane resources.
• Database users can be scoped with built-in roles, specific privileges, and custom roles.
• Atlas supports multiple database-user auth methods, including SCRAM and X.509, with environment-sensitive guidance in the docs.

Practical security defaults

1. Prefer private networking for higher-security production environments.
2. Prefer service accounts for Atlas control-plane automation.
3. Keep API and cluster network allow-lists narrow and time-bound where possible.
4. Use the minimum project/organization role that can perform the task.
5. Keep the distinction between control-plane auth and data-plane auth explicit in design docs and code.

MongoDB coding standards that still apply on Atlas

Schema design

• Model data around application access patterns.
• Keep data that is accessed together together.
• Use embedding when it reduces joins and multi-document coordination.
• Use referencing when data has a different access cadence, lifecycle, or cardinality profile.

Query and index design

• Add indexes for repeated query shapes.
• Remember every additional index raises write cost.
• Atlas Performance Advisor is useful, but recommended indexes still need human judgment about workload frequency and write tradeoffs.
• Large arrays and $lookup-heavy designs are called out in Atlas docs as common sources of slow-query pain.

Write and transaction design

• Single-document operations are atomic; many practical designs should exploit that instead of defaulting to transactions.
• Include expected current state in update filters or use intent-specific operators like $inc where concurrency matters.
• Use transactions only when multi-document, multi-collection, or cross-shard atomicity is actually required.

Driver usage

• Production application code should usually target an official MongoDB driver.
• Atlas expertise includes choosing the right driver-level behavior for the language/runtime, not only knowing Atlas admin features.

Search and vector-search standards

Atlas Search

• Atlas Search is the embedded full-text and relevance-search system for Atlas.
• Search queries are expressed through aggregation pipeline stages such as $search and $searchMeta.
• Atlas Search supports analyzers, autocomplete, pagination, faceting, scoring, static mappings, and dynamic mappings.
• Use Atlas Search when the workload is fundamentally about relevance-based text retrieval.

Atlas Vector Search

• Atlas Vector Search is the semantic/vector retrieval surface for Atlas.
• Use it for semantic search, hybrid search, and RAG/agentic retrieval patterns.
• Atlas docs call out ANN and ENN availability by MongoDB version; do not give vector-search guidance without checking version support.
• Automated Embedding uses the autoEmbed index field type with built-in Voyage AI models (voyage-4-large, voyage-4, voyage-4-lite, voyage-code-3) to generate embeddings on insert, update, and query without an external pipeline. Public preview as of May 2026. See the refresh section below for pricing and model details.

Backup, reliability, and disaster-recovery standards

• Atlas Cloud Backup uses the cloud provider’s native snapshot functionality and inherits provider redundancy guarantees.
• Atlas supports multi-region snapshot distribution for added redundancy and region-failure recovery posture.
• Backup restore/admin operations require the appropriate project backup roles; org-level access alone is not sufficient until explicitly added to the project.
• Topology matters: Atlas docs note restore caveats for sharded clusters after shard-count changes.

Monitoring, alerts, and capacity standards

Alerts

• Atlas alerts are configured around conditions, thresholds, notification methods, and lifecycle management.
• Alert policies can exist at the organization or project level.
• Atlas supports acknowledge/unacknowledge, disable/enable, and delete flows for alerts.

Metrics

• Atlas collects metrics across servers, databases, and MongoDB processes.
• High-signal metrics called out in the docs include connections, disk IOPS, disk usage, query targeting, and normalized system CPU.
• Atlas exposes real-time metrics and Atlas Search metrics in addition to cluster views.

Capacity and limits

• Atlas limits around connections, shards, nodes, and topology are design inputs, not just operational trivia.
• Atlas docs explicitly recommend connection pooling, application tuning, autoscaling, and tier scaling when nearing connection limits.

High-value method inventory

This is the condensed inventory to keep handy. For exhaustive method lists, use the linked references directly.

Atlas control-plane methods

Atlas-connected application/data methods

Practical defaults for future Atlas coding/review tasks

1. Start with which Atlas surface the task belongs to: UI, Admin API, CLI, Terraform, AKO, or application driver.
2. Separate Atlas control-plane actions from MongoDB data-plane actions before proposing code.
3. For application design, start with access patterns, then schema, then indexes, then aggregation/search/vector shape.
4. For operations, review alerts + metrics + limits + backup posture together.
5. For security, review network path + identity model + required role + auth method together.
6. For AI/search work, choose explicitly between Atlas Search, Atlas Vector Search, or hybrid patterns.

Recent platform changes (2025-2026 refresh)

This section captures major Atlas and MongoDB platform changes since late 2024. Last refreshed: 2026-05-25.

MongoDB server versions on Atlas

• MongoDB 8.0 (GA October 2024): 32% query throughput improvement, 56% faster bulk writes, 200% faster time-series aggregations, 50x faster data distribution for sharding at 50% lower cost. Introduced default maximum time limits for queries and the ability to reject recurring problem queries. https://www.mongodb.com/docs/manual/release-notes/8.0/
• MongoDB 8.2 (2025): Public preview of enhanced Queryable Encryption (prefix, suffix, substring queries on encrypted fields), $currentDate in aggregate(), standardized spill-to-disk metrics in explain output. https://www.mongodb.com/docs/manual/release-notes/8.2/
• MongoDB 8.3 (May 2026): ~45% more reads and ~35% more writes vs 8.0, sub-100ms retrieval targets for agent workloads, new $hash and $hexHash aggregation expressions (MD5, SHA-256, XXH64), arrayIndexAs field in $map/$filter/$reduce, removeShard deprecated in favor of four new drain/removal commands, security hardening and native type-coercion expressions. https://www.mongodb.com/docs/manual/release-notes/8.3/

Flex clusters (replaces Shared and Serverless tiers)

• Atlas Flex tier is the unified replacement for M2, M5, and Serverless instances. It combines the best of Shared and Serverless into a single offering with dynamic scaling.
• As of March 2025, Serverless instances are no longer supported; existing instances were migrated to Free, Flex, or Dedicated clusters.
• As of May 2025, all M2/M5 clusters have been auto-migrated to Flex.
• As of January 2026, the old createGroupCluster (M2/M5) and createGroupServerlessInstance API endpoints only support Flex clusters.
• Flex includes 100 ops/sec and 5 GB storage by default, scales to 500 ops/sec dynamically, $8 base + usage-based billing capped at $30/month. https://www.mongodb.com/docs/atlas/manage-flex-clusters/
• Flex key limits: 500 connections max, 5 GB storage hard cap (no auto-expand), 500 collections max, 100 databases max, MongoDB 8.0 minimum (auto-upgrade only).
• Flex does NOT support: Private Endpoints (no PrivateLink/VPC peering), Continuous Backup/PITR (daily snapshot only), BYOK encryption at rest, Database Auditing, Performance Advisor, Rolling index builds, allowDiskUse, server-side JS.
• Flex DOES support (unlike old M2/M5): Atlas Search, Atlas Vector Search, Change Streams, Triggers — but Vector Search shares resources with mongod on Flex; upgrade to M10+ with dedicated Search Nodes before production Vector Search.
• Migration is one-way: Flex → dedicated is supported (downtime required); dedicated → Flex downgrade is NOT supported. Download Flex snapshots before upgrading as they do not transfer to dedicated clusters.
• For full Flex decision matrix, pricing breakdown, and tooling migration, see mongodb-atlas-flex-serverless skill.

Atlas Stream Processing (GA)

• Atlas Stream Processing reached general availability as of March 2025.
• Enables continuous stream processing pipelines over Atlas data using aggregation-pipeline syntax.
• Supports emitting to Time Series Collections, Kafka headers, and multiple tiers (SP10 for low-traffic, SP30 for production).
• Available on AWS and Azure across global regions. https://www.mongodb.com/docs/atlas/atlas-stream-processing/

Voyage AI acquisition and Automated Embedding

• MongoDB acquired Voyage AI in February 2025 (~$220M) to embed high-accuracy embedding models directly into Atlas Vector Search.
• Automated Embedding (public preview May 2026) uses the autoEmbed index field type to automatically generate Voyage AI vector embeddings on insert, update, and query – no external pipeline needed.
• Available models: voyage-4-large, voyage-4, voyage-4-lite, voyage-code-3.
• Pricing: per million tokens ($0.12 large / $0.06 standard / $0.02 lite); first 200M tokens free per account; Batch API gives 33% discount. https://www.mongodb.com/docs/atlas/atlas-vector-search/

Lexical Prefilters for Vector Search

• Lexical Prefilters allow advanced text and geo analysis filters (fuzzy search, phrase matching, wildcards, geoWithin) as prefilters before vector similarity search.
• Unlike standard $vectorSearch filters (equals, range, exists), lexical prefilters use full analyzed-text capabilities from Atlas Search operators.
• Create a $search index with vector type fields and use $search.vectorSearch in aggregation pipelines. https://www.mongodb.com/company/blog/product-release-announcements/semantic-power-lexical-precision-advanced-filtering-for-vector-search

Hybrid Search with $rankFusion

• The $rankFusion aggregation operator merges and re-ranks results from multiple search pipelines (full-text + vector).
• Requires MongoDB 8.1+ on Atlas.
• Enables true hybrid search combining keyword precision with semantic intelligence in a single query. https://www.mongodb.com/docs/atlas/atlas-vector-search/hybrid-search/

Search Nodes (dedicated search infrastructure)

• Search Nodes are generally available on AWS, Google Cloud, and Azure for both development and production deployments.
• Provide dedicated infrastructure for Atlas Search and Vector Search, independent of database compute, with up to 60% query-time reduction.
• Multi-region Search Nodes are available in preview for multi-region and multi-cloud clusters. https://www.mongodb.com/docs/atlas/atlas-search/

Queryable Encryption enhancements

• Equality and range queries on encrypted fields are GA and production-ready at no additional cost on Atlas, Enterprise Advanced, and Community Edition.
• Prefix, suffix, and substring queries on encrypted string fields are in public preview starting MongoDB 8.2. https://www.mongodb.com/docs/manual/core/queryable-encryption/

Terraform Provider 2.0

• MongoDB Atlas Terraform Provider 2.0 shipped in 2025 with semantic versioning, no-breaking-change guarantees in minor/patch releases, eliminated hanging timeouts, and simplified advanced-cluster migrations.
• Migration required from 1.x; see the 2.0.0 Upgrade Guide.
• Atlas Architecture Center examples now target Provider 2.x. https://www.mongodb.com/products/updates/terraform-mongodb-atlas-provider-2-0-now-available/

Atlas Admin API standardized rate limiting

• Standardized rate limiting for the Atlas Admin API v2 became GA in March 2026, using a token-bucket algorithm.
• Automation and integration code should handle 429 Too Many Requests responses and respect Retry-After headers. https://www.mongodb.com/company/blog/product-release-announcements/introducing-standardized-atlas-admin-api-rate-limiting

Atlas Architecture Center compliance additions

• PCI DSS Compliance page added February 2026.
• HIPAA Compliance page added February 2026.
• Multi-region opinionated guidance, Reliability section, and Operational Readiness Checklist added August 2025. https://www.mongodb.com/docs/atlas/architecture/current/changelog/

Atlas CLI updates (2025-2026)

• atlas api subcommand reached GA in October 2025.
• TLS 1.3 support added for atlas api clusters commands (December 2025).
• atlas api aiModelRateLimits resetModelRateLimit command added April 2026. https://www.mongodb.com/docs/atlas/cli/current/atlas-cli-changelog/

AI and agentic positioning

• MongoDB is positioning Atlas as a converged datastore for agentic AI: operational data + vector search + stream processing + agent memory in one platform.
• First-class integrations with LangGraph.js (long-term memory store, GA), and major agent frameworks.
• Atlas AI Integrations documentation consolidates RAG, agent, and embedding patterns. https://www.mongodb.com/docs/atlas/ai-integrations/

Known ambiguities and guardrails

• “All Atlas methods” is too large for a single static file. Use this context as the condensed expert map, then jump to the linked Admin API, CLI, driver, and Atlas feature references for exact syntax and complete inventories.
• Atlas docs are versioned and living. For vector search, API behavior, CLI commands, and limits, always confirm the current version or page timestamp when precision matters.
• Atlas advice often differs by cluster tier, deployment topology, cloud provider, and MongoDB version. Good answers should say which of those variables matter.

See Also

For deep Atlas sub-areas (Azure, GCP, multicloud, Search, Vector Search, and 22 more), use the Sub-skill routing table above and read the matching references/…md file — those topics are now consolidated into this hub.

Peer hubs to hand off to:

• [[mongodb-expert]] — data-plane query, index, schema, aggregation, and storage-engine work
• [[atlas-diagnostics-expert]] — live cluster diagnostics, performance, monitoring, and capacity
• [[mongodb-operations-expert]] — backups, DR, Ops Manager, migration, mongosync, security architecture, encryption, compliance, connectors, and cost
• [[mongodb-kb]] — MongoDB knowledge-base article lookup

Cross-hub map — where every MongoDB topic lives

All MongoDB knowledge is split across four hubs (plus mongodb-kb for KB-article lookups and 10gen for repo install/run). If a task’s deep material is not in this hub’s Sub-skill routing table, it is a reference file under a sibling hub — activate that hub or Read its references/<name>.md directly.

High-overlap routing notes:

• Performance symptom triage (high CPU, cache pressure, slow queries, latency spikes) starts at atlas-diagnostics-expert, but storage-engine root-cause internals (WiredTiger cache fill / dirty trigger / eviction threads / reconciliation / checkpoints) are owned by mongodb-expert — cross-load mongodb-expert/references/mongodb-wiredtiger-internals.md (and mongodb-wiredtiger.md) for depth.
• Migration symptoms vs migration execution: live-cluster diagnosis → atlas-diagnostics-expert; the migration/mongosync runbook → mongodb-operations-expert.
• Atlas Search/Vector query syntax & index design → mongodb-atlas-expert; the slowness triage of a running search → atlas-diagnostics-expert.