# How to Deploy Your First AI Agent on MongoDB Atlas Agent Engine, Powered by Amazon Bedrock

DevFeed: [How to Deploy Your First AI Agent on MongoDB Atlas Agent Engine, Powered by Amazon Bedrock](<https://devfeed.tech/articles/how-to-deploy-your-first-ai-agent-on-mongodb-atlas-agent-engine-powered-by-amazon-bedrock-121432.md>)

Original publisher: [Read original article](<https://blog.devops.dev/how-to-deploy-your-first-ai-agent-on-mongodb-atlas-agent-engine-powered-by-amazon-bedrock-85a7ca4628b6?source=rss----33f8b2d9a328---4>)

Author: Carlos Biagolini

Published: 2026-11-06T09:47:13Z

Content type: tutorial

Language: en

Sources: [DevOps.dev](<https://devfeed.tech/sources/devops-dev.md>)

Topics: [agent observability](<https://devfeed.tech/topics/agent-observability.md>), [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [Langgraph](<https://devfeed.tech/topics/langgraph.md>), [Amazon Bedrock AgentCore](<https://devfeed.tech/topics/amazon-bedrock-agentcore.md>), [AI Chat](<https://devfeed.tech/topics/ai-chat.md>), [google-adk](<https://devfeed.tech/topics/google-adk.md>), [voyage-ai](<https://devfeed.tech/topics/voyage-ai.md>), [vercel ai sdk](<https://devfeed.tech/topics/vercel-ai-sdk.md>)

Tags: [2026](<https://devfeed.tech/tags/2026.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agentic-ai](<https://devfeed.tech/tags/agentic-ai.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agent](<https://devfeed.tech/tags/ai-agent.md>), [amazon-bedrock](<https://devfeed.tech/tags/amazon-bedrock.md>), [atlas](<https://devfeed.tech/tags/atlas.md>), [atlas-agent-engine](<https://devfeed.tech/tags/atlas-agent-engine.md>), [cli](<https://devfeed.tech/tags/cli.md>), [cloud](<https://devfeed.tech/tags/cloud.md>), [conversations](<https://devfeed.tech/tags/conversations.md>), [data](<https://devfeed.tech/tags/data.md>), [deploy](<https://devfeed.tech/tags/deploy.md>), [framework](<https://devfeed.tech/tags/framework.md>), [frameworks](<https://devfeed.tech/tags/frameworks.md>), [google-adk](<https://devfeed.tech/tags/google-adk.md>), [governance](<https://devfeed.tech/tags/governance.md>), [governance-layer](<https://devfeed.tech/tags/governance-layer.md>), [how-to](<https://devfeed.tech/tags/how-to.md>), [mongodb](<https://devfeed.tech/tags/mongodb.md>), [mongodb-atlas](<https://devfeed.tech/tags/mongodb-atlas.md>)

## AI overview

The walkthrough builds a LangGraph Python agent that calls Claude through Amazon Bedrock, tests it locally, links it to MongoDB Atlas, and deploys it to Atlas Agent Engine. It uses a Flex cluster because Free clusters are unsupported; guided setup creates a database user, stores the connection string as a platform secret, and adds Agent Engine IPs to the cluster access list. The deployed agent also needs its Bedrock key stored as a secret. The platform runs an orchestration engine with separate agent and tool sandboxes, while execution state, checkpoints, and history reside in Atlas. The example leaves memory off and does not configure custom tools or guardrails. The article flags the tutorial's broad `readWriteAnyDatabase` user, project-wide default secret access, session capacity limits, and public-preview status; it recommends a dedicated project for experimentation. Its comparison with Amazon Bedrock AgentCore centers on state location and assembly: AgentCore uses managed memory and session services as modular components, while Atlas Agent Engine stores state in the user's Atlas cluster and bundles memory, retrieval, and governance.

## Source excerpt

Introduction Building an AI agent that works in a demo is quick. Running it in production is a different problem: the agent needs memory that survives between conversations, accurate retrieval over company data, controlled access to tools and networks, and a trace of everything it did. Teams usually solve this by stitching together a vector store, a cache, a checkpointer, a secrets manager, a tracing tool, and a custom runtime, and that stack tends to break every time the model or the framework changes. MongoDB Atlas Agent Engine, which MongoDB launched into public preview in September 2026, is a managed platform built to replace that stack. MongoDB describes it as a unified execution, memory, and governance layer for production agents. It runs your agent code in isolated agent and tool sandboxes coordinated by an orchestration engine, and adds the capabilities that agents usually lack: Long-term memory, with semantic, episodic, taxonomic, and procedural memory managed by the platform. Retrieval backed by Atlas Search, Atlas Vector Search, and Voyage AI embedding and reranking models. Governance enforced at runtime: every tool call runs with a real identity and least privilege, outbound network access is denied unless declared, and policies are defined once at the organization level. Observability without instrumentation, with every model call, tool call, and memory access traced. MongoDB frames the differentiator against competing platforms around where the agent runs and what it avoids locking in. The company describes a common tradeoff: teams either adopt a single vendor's agent runtime, which ties them to that vendor's models and cloud, or assemble an open source framework themselves, which leaves memory and governance as their own problem. Atlas Agent Engine is positioned between the two, with a few properties worth checking when you evaluate it: Neutral across models and frameworks. The CLI offers LangGraph and Google ADK templates in Python or TypeScript, and