Web tools let Kiro access current information from the internet in real-time, grounding responses with up-to-date search results and page content. These tools are designed not to reproduce meaningful chunks of text, and they work with publicly accessible webpages.
Web tools are available across all surfaces (IDE, CLI, Web, and Mobile).
| Tool | Description |
|---|---|
web_search | Search the web for current information |
web_fetch | Fetch and extract text from a URL |
In the following demo, the agent automatically uses web_search to return up-to-date results for a query, then uses web_fetch to retrieve the latest contents at a specific URL:
Searches the web and returns relevant results. The agent uses this automatically when you ask about current events, recent changes, or topics beyond the model's training data.
> What is the latest on EC2 instances? Searching the web for: AWS EC2 instances latest 2025 (using tool: web_search) ✓ Found 10 search results
Fetches and extracts text content from a specific URL. Supports three modes for managing context efficiently.
| Mode | Behavior | Use case |
|---|---|---|
selective (default) | Returns sentences around search term matches | Targeted extraction |
truncated | First 8,000 characters | Quick preview |
full | Complete content (up to 10 MB) | Comprehensive analysis |
> Get installation info from https://kiro.dev/blog/introducing-kiro-cli/ Fetching content from: https://kiro.dev/blog/introducing-kiro-cli/ (searching for: installation install getting started) [mode: selective] ✓ Fetched 7909 bytes (selective) from URL
When no specific content is requested, the tool defaults to truncated mode (first 8 KB) to manage context.
| Constraint | Value |
|---|---|
| Maximum page size | 10 MB |
| Timeout | 30 seconds |
| Maximum redirects | 10 |
| Content type | text/HTML only |
| Retries | 3 automatic attempts |
Administrators using IAM Identity Center can disable web tools for their organization:
See Enterprise governance - Web tools for details.
To prevent an agent from using web tools, exclude the web tag:
{ "tools": ["read", "write", "shell"], "excludedTools": ["web"] }
Or use permissions to deny web access:
rules: - capability: web_fetch effect: deny - capability: web_search effect: deny
Web tools