Web Search Engine

Find missing information on the web.

This feature enables the agent to search the internet for information that may not be available in the local codebase or documentation. It can help answer questions about current APIs, libraries, frameworks, or any other topics that require up-to-date information from the web.

Supported agents:

• Claude Code
• Codex
• Cursor

Not supported agents:

• Kilo Code
• Junie
• OpenCode

Not verified yet:

None

Fetch Data

Get real-time answers from any website.

You cannot just use curl or wget to fetch data from the web because most websites return a lot of irrelevant content, such as HTML, CSS, and JavaScript. The built-in tool should be able to download the data and extract the useful information, filtering out any unnecessary clutter or noise.

Supported agents:

• Claude Code
• OpenCode https://opencode.ai/docs/tools/#webfetch

Not supported agents:

• Codex
• Cursor
• Kilo Code
• Junie

Not verified yet:

None

Browser

Browse the web just like a human.

If you’re a web developer or want to automate tasks, you can use the built-in browser feature. This approach is better than MCP browser integration because coding clients can target common actions directly—for example, “take a screenshot of the page [abc].”

Supported agents:

• Cursor (partial support). You can call a browser only in plan and agent modes. Ask and debug modes are not supported the browser tool.

Not supported agents:

• Codex
• Kilo Code (The developer declares the support here https://kilo.ai/docs/features/browser-use) but I failed to connect Chrome instances to the agent)
• Junie
• OpenCode

Not verified yet:

• Claude Code

Linters

Code fixes, automated by agents.

Linters look for mistakes and style problems in your code. Agents can use these tools to find and fix issues without you doing it by hand or using extra tokens.

You could make your own skill (see partial/skills page), but that uses more context and tokens. Agents can get a list of errors and fix them directly, without needing an AI model. This way is faster, cheaper, and more reliable.

Supported agents:

• Cursor (partial support): You can install plugins for your linter—such as ESLint, TypeScript, etc.—and Cursor will detect errors directly, without consuming tokens. However, it cannot run --autofix to automatically fix issues; instead, it uses the selected model to resolve the errors.
• OpenCode (https://opencode.ai/docs/lsp/)

Not supported agents:

• Codex
• Kilo Code
• Junie

Not verified yet:

• Claude Code