Agent tools

Available tools#

Search files and folders locates code by name or content across the repository. This usually runs first on any request that names something without giving a path.

Read files pulls file contents into context. The agent reads before editing rather than guessing at what a file contains.

Edit files applies changes. Every edit appears in the diff view for review.

Run shell commands executes terminal commands: tests, builds, linters, package installs, and git operations.

Web fetches documentation and other pages, which matters when working against a library whose API changed after the model's training cutoff.

Browser drives a browser session so the agent can check rendered output rather than reasoning about markup.

Fetch rules pulls in the rules relevant to the current task.

Image generation produces images where a task calls for one.

Ask questions lets the agent stop and ask you instead of guessing on an ambiguous requirement.

Tool calls are unlimited per task#

A single task can involve any number of tool calls. A well-scoped request produces a short, legible sequence. A vague one produces a long exploratory search that costs time and usage.

Extending the tool set#

MCP adds tools beyond the built-in set, connecting the agent to databases, internal APIs, and third-party services.

Related#

• Agent overview — how tools fit with instructions and the model
• MCP — connect external systems