I Tried Out the Web Access Feature of Kiro-CLI (v1.21.0)

On November 25, 2025, version 1.21.0 of the AI assistant "Kiro-CLI" was released.

This update adds support for the web_fetch and web_search tools, enabling Kiro-CLI to access content on the internet.

https://kiro.dev/changelog/web-search-and-fetch-for-kiro-cli/

https://kiro.dev/docs/cli/reference/built-in-tools/#web-search-and-fetch

In this article, I tested on a real machine how these differ from the existing curl, as well as the behavior and limitations of the search feature.

Test Environment

For the test, I prepared the following setup.

• Client: Amazon Linux 2023 (Kiro-CLI v1.21.0)
  • Updated with kiro-cli update

$ kiro-cli version
kiro-cli 1.21.0

• Web server: Amazon Linux 2023 (EC2)
  * Ran a simple HTTP server (port 3000) using the nc command

# Simple web server (returns "OK" on port 3000)
while true; do { echo -e "HTTP/1.1 200 OK\r\nContent-Length: 2\r\n\r\nOK"; } | nc -l -p 3000 2>&1; done

1. web_fetch: Retrieving Content by Specifying a URL

I tested the web_fetch tool, which retrieves the contents of a specified URL.

1-1. Basic Behavior and Differences from curl (shell tool)

Kiro-CLI has always been able to run curl via the shell tool.

To determine how to choose between that and the newly added web_fetch, I instructed Kiro to access the simple web server with both tools and had Kiro itself analyze the differences.

Prompt:

Display the content of http://52.88.xxx.xxx:3000 by using tool: web_fetch and by running curl. Also tell me the differences.

Results:

1. web_fetch: Fetching content from... → Retrieved as formatted HTML content
2. curl (shell): curl http://... → Retrieved the raw response

Both accessed the server successfully, and Kiro's answer on when to use each was as follows.

web_fetch is suitable when you want the AI to read the contents (text information), while curl can still be used for testing the communication itself or for processing that involves authentication.

1-2. Verifying the User-Agent

I checked the request headers that reached the web server during web_fetch.

GET / HTTP/1.1
accept: */*
user-agent: Kiro-CLI
accept-encoding: gzip
host: 52.88.xxx.xxx:3000

When using the curl command, the User-Agent is curl/x.x.x, but with web_fetch it is user-agent: Kiro-CLI, confirming that it is a different client.

1-3. Extraction via Selective Mode (robots.txt)

Using DevelopersIO's robots.txt as the target, I instructed it to extract only the "Allow" lines.

Execution log:

Fetching content from: https://dev.classmethod.jp/robots.txt (searching for: Allow) [mode: selective] (using tool: web_fetch)
...
> はい、抽出できました。selectiveモードで"Allow"をキーワードに検索した結果、以下のAllowパスが含まれています:

- Allow: /
- Allow: /sitemap.xml
- Allow: /sitemaps/*
- Allow: /feed/
- Allow: /en/feed/

The log displayed [mode: selective], and only the lines related to the specified keyword were extracted rather than the whole file. I confirmed that this is useful when you want to reference only a specific section of a document.

1-4. Specifications and Limitations

From the documentation and this test, I confirmed the following specifications and limitations.

• Supported content: text/html only (image and PDF parsing not possible)
• File size: up to 10MB
• Timeout: 30 seconds
• Authentication: Pages requiring Basic authentication or login are not supported

web_search

Next, I tested the web_search tool, which searches for information on the web like Google search.

Searching Recent Information (November 2025)

I tried asking about recent information that should not be in the AI's training data (at the time of testing: November 2025).

Prompt:

Tell me the new AWS services released in November 2025, with service names and announcement dates

Answer:

(Search results)
Here is a summary of the major AWS services and features released in November 2025, by service name and>announcement date:

November 19

• Amazon Connect: Conversational analytics (for voice and chatbots)
• Amazon Bedrock: Support for new regions (Africa, Canada, Mexico, Middle East)
• Amazon OpenSearch Serverless: Backup and restore feature (AWS Management Console support)
• AWS Secrets Manager: Managed external secrets feature
• Amazon ECS Managed Instances: Scale-in delay setting feature
• AWS PrivateLink: Cross-region connectivity support
• Amazon ECR: Archive storage class

November 21

• Amazon EMR Serverless: Apache Spark 4.0.1 (preview)
• Amazon Athena for Apache Spark: SageMaker Notebooks integration
• AWS Payments Cryptography: Post-quantum cryptography support
• AWS Device Farm: Managed Appium endpoint
• Amazon EKS/ECS: Fully managed MCP server (preview)
• Amazon ECS Express Mode: New feature release

November 24

• Amazon OpenSearch Service: Enhanced PPL log analytics
• Amazon CloudFront: Mutual TLS authentication support
• Amazon EC2: Interruptible Capacity Reservations
• Claude Opus 4.5: Available on Amazon Bedrock
• Amazon Aurora PostgreSQL: Dynamic data masking feature
• Amazon SageMaker HyperPod: NVIDIA MIG support

References:
[1] Daily AWS Headlines - https://www.daily-aws.com/headlines/20251119/
[2] Daily AWS Headlines - https://www.daily-aws.com/headlines/20251121/
[3] Daily AWS Headlines - https://www.daily-aws.com/headlines/20251124/

The list included information from just a few days before the time of writing (2025/11/27). Source URLs were also clearly provided.

Information from 2014 (the "Past")

For comparison, I also checked the accuracy of information from 11 years ago.

Prompt:

Tell me the new AWS services released in November 2014, with service names and announcement dates

Answer:

(Search results)
The major AWS services released in November 2014 are as follows:

November 12

• AWS Key Management Service (KMS): Encryption key management service
• AWS CodeDeploy: Automated code deployment service

November 13 (announced)

• AWS Lambda: Event-driven compute service

November 14 (AWS re:Invent 2014 Day 2 Keynote)

• Amazon EC2 Container Service (ECS): Docker container management service

Announced in preview (scheduled for release in early 2015)

• Amazon RDS for Aurora: MySQL-compatible relational database
• AWS Config: Resource configuration management service
• AWS CodeCommit: Code management service
• AWS CodePipeline: CI/CD pipeline service
• AWS Service Catalog: Service catalog management

Other feature additions

• Amazon S3 Event Notifications: S3 event notification feature (SNS/SQS/Lambda support)

References:
[1] AWS Developer Tools Blog - https://aws.amazon.com/blogs/developer/aws-reinvent-2014-
recap-2/

It answered with information from the year in which Lambda, Aurora, and others were announced.

Search Refusal

This search tool appears to be specialized for "technical research."
When I asked a question unrelated to technology (e.g., geographic information), it was clearly refused.

Prompt:

Q: Tell me a web page that introduces the westernmost point of mainland Japan

Answer:

I'm sorry, but that question is not related to AWS, cloud technology, or development, so I cannot use the web_search tool.

Use as a general-purpose search engine is restricted, and its positioning as a "technical assistant for developers" appears to be controlled at the system prompt level.

Summary

I summarized the test results for the web access features supported in Kiro-CLI v1.21.0, compared with the existing curl.

Kiro-CLI automatically selects tools based on the user's instructions, but by understanding the strengths and constraints of each tool, users who are developers can bring out the full potential of the AI development assistant.

Keeping the following characteristics in mind and writing appropriate prompts is the key to getting faster and more accurate results.

I hope you can make use of these characteristics for more advanced use of Kiro-CLI.