5 result(s)
The post announces migrating VeraResearch into first-class MCP tool support inside Strands, emphasizing a shift to let teams ‘manage the complexity’ by providing native, supported tooling and integrations that simplify workflows and reduce accidental overhead.
This article explains how to build audit logging for multi-platform AI bots using Python and AWS CloudWatch to answer questions like “who is actually using our bots today?”. It covers architecture, event schema, ingestion, storage, querying, privacy considerations, and practical implementation tips to capture, ship, and analyze bot usage across channels.
The article explains how to teach Agentic Strands agents to assume cross-account AWS IAM roles so they can read resources (for example, to answer operational questions like which VMs are running high CPU or low on disk). It covers the motivation, high-level workflow (AssumeRole via STS), and practical/security considerations for enabling safe cross-account access by autonomous agents.
The article explains why multiple MCP servers often all register a tool named “search”, causing name collisions when a client aggregates tools from more than one server, and gives short-term and long-term fixes (namespacing, unique IDs, client-side resolution changes, or upstream fixes).
The article addresses a common enterprise problem with AWS Bedrock: unknown consumption of AI tokens. It proposes adding explicit bot/user context to model invocation logs so teams can attribute usage, monitor costs, enforce policies, and investigate behavior. The approach uses structured metadata injected at the application or gateway layer and captured in Bedrock invocation logs and downstream observability systems.