This guide explains how to effectively use Model Context Protocol (MCP) tools with Kiro to enhance your productivity and access specialized capabilities.
Interacting with MCP Tools
Once you've configured MCP servers, you can interact with their tools in several ways:
Direct Questions
The simplest way to use MCP tools is to ask questions related to the server's domain:
Tell me about Amazon Bedrock
How do I configure S3 bucket policies?
Kiro automatically selects the appropriate MCP tool based on your question.
Specific Tool Requests
You can request specific MCP tools by describing what you want to do:
Search AWS documentation for information about ECS task definitions
Get recommendations for AWS CloudFormation best practices
Explicit Context
For more control, provide explicit context to the tool picker:
#[aws-docs] search_documentation Tell me about AWS Lambda
This format specifies both the server (aws-docs
) and the tool (search_documentation
).
MCP Tools Panel
The Kiro panel includes an MCP servers tab that provides:
- A list of all configured MCP servers
- Connection status indicators
- Quick access to server tools
To use this feature:
- Select the Kiro icon in the activity bar
- Navigate to the MCP servers tab
- Click any tool name to insert a placeholder prompt in the chat
Tool Approval Process
When Kiro wants to use an MCP tool, it requests your approval first:
- You'll see a prompt describing the tool and its purpose
- Review the tool details and parameters
- Click "Approve" to allow the tool to run, or "Deny" to prevent it
Auto-approving Trusted Tools
To avoid repeated approval prompts for tools you trust:
- Edit your MCP configuration file
- Add tool names to the
autoApprove
array:
{ "mcpServers": { "aws-docs": { "autoApprove": [ "mcp_aws_docs_search_documentation", "mcp_aws_docs_read_documentation" ] } } }
- Restart Kiro to apply changes
Examples by Server Type
AWS Documentation Server
Searching Documentation
Search AWS documentation for S3 bucket versioning
This uses the mcp_aws_docs_search_documentation
tool to find relevant AWS documentation.
Reading Documentation
Read the AWS Lambda function URLs documentation
This uses the mcp_aws_docs_read_documentation
tool to retrieve and display documentation content.
Getting Recommendations
Find related content to AWS ECS task definitions
This uses the mcp_aws_docs_recommend
tool to suggest related documentation.
GitHub MCP Server
Repository Information
Show me information about the tensorflow/tensorflow repository
This retrieves details about the specified GitHub repository.
Code Search
Find examples of React hooks in facebook/react
This searches for code matching your query in the specified repository.
Issue Management
Create an issue in my repository about the login bug
This helps you create a new GitHub issue with appropriate details.
Advanced Usage Techniques
Chaining MCP Tools
You can use multiple MCP tools in sequence for complex tasks:
First search AWS documentation for ECS task definitions, then find related content about service discovery
Combining with Local Context
MCP tools work best when combined with your local context:
Based on my Terraform code, help me optimize my AWS Lambda configuration using best practices from AWS documentation
Using MCP Tools in Specs
You can use MCP tools within Kiro Specs to enhance your development workflow:
In the implementation phase, use AWS documentation to ensure our S3 bucket configuration follows best practices
Troubleshooting Tool Usage
If you encounter issues when using MCP tools:
Tool Not Responding
- Check the MCP server status in the Kiro panel
- Review the MCP logs for error messages
- Restart the MCP server if necessary
Incorrect Results
- Try rephrasing your request to be more specific
- Check that you're using the appropriate tool for your task
- Verify that the MCP server has the necessary permissions
Tool Not Available
- Ensure the MCP server is properly configured
- Check that the server is running and connected
- Verify that you have the necessary permissions to use the tool
Best Practices
- Be specific in your requests to get the most relevant results
- Start with direct questions before using explicit tool references
- Auto-approve only tools you trust and use frequently
- Combine MCP tools with local context for best results
- Check tool parameters before approval to ensure they're correct
For information on available MCP servers and their tools, see the MCP Servers page.
Tools