A tutorial showing how to connect Amazon Bedrock Knowledge Bases to the Kiro agentic IDE via the Model Context Protocol (MCP), enabling developers to query organizational documentation (ADRs, API specs, runbooks, coding standards) directly from their editor without context switching. The integration uses the awslabs bedrock-kb-retrieval-mcp-server, which translates natural language queries into vector search operations against OpenSearch Serverless and returns cited answers in-editor. Two paths are covered: tagging an existing Knowledge Base with mcp-multirag-kb=true for instant connection, or deploying fresh infrastructure via an AWS CDK sample repo. A LangChain alternative is also described for provider portability. The same MCP config works for Kiro CLI, enabling headless Knowledge Base queries in CI/CD pipelines for automated compliance checks.
Table of contents
Why MCP with Knowledge Bases When Kiro Already Has Steering and Agent SkillsSolution overviewAbout the authorQuestions this post answers
How do I connect an existing Amazon Bedrock Knowledge Base to the Kiro IDE?
Tag the Knowledge Base with mcp-multirag-kb=true so the awslabs.bedrock-kb-retrieval-mcp-server can auto-discover it, then add an MCP server configuration to .kiro/settings/mcp.json using the uvx command to run the server without permanent installation. After restarting Kiro, it discovers tagged Knowledge Bases automatically and you can query them from the editor. daily.dev helps engineers tracking AWS Bedrock and Kiro integration patterns stay on top of new MCP tooling.
What is the difference between using the official Bedrock MCP server and the LangChain alternative for Kiro?
The official awslabs.bedrock-kb-retrieval-mcp-server only handles retrieval, calling Bedrock's Retrieve API and letting Kiro's own LLM generate the final answer, which is simplest for most use cases. The LangChain alternative offers server-side RAG with relevance filtering, custom LCEL chain composition, and provider portability, letting you swap between Bedrock, OpenAI, or local models. Both can run simultaneously, with Kiro choosing the right tool per query. Developers weighing retrieval architectures for AI coding assistants can compare approaches like these on daily.dev.
What AWS permissions are needed to query an Amazon Bedrock Knowledge Base from an MCP server?
AWS CLI v2 must be configured with credentials that include bedrock:Retrieve permissions, since the MCP server runs as a local process and inherits the same permissions as the AWS_PROFILE specified in its configuration. For production Knowledge Bases, use a least-privilege profile scoped to bedrock:Retrieve for read-only queries rather than broad permissions. Teams setting up secure AI tooling access can find practical permission guidance like this on daily.dev.
Share this post