Tools

A tool is an action the agent can take on your behalf. On its own, a model can only produce text; tools are what let gdc actually read your files, change code, and run commands. When the model decides it needs to act, it requests a tool, gdc checks it against your permissions, runs it, and hands the result back.

You don't call tools yourself — the agent chooses them as it works. This page explains what's available so you know what gdc can and can't do.

Built-in tools

Every session has these tools available:

Tool	What it does
Read	Read a file (or part of one).
Write	Create a new file or overwrite an existing one.
Edit	Make a precise change to an existing file.
Bash	Run a shell command.
Glob	Find files by name pattern.
Grep	Search file contents.
Web fetch	Retrieve a web page (for documentation, references).

Reading and searching are safe and generally run without interruption. Writing, editing, and running commands can change your system, so they're governed by permissions and gdc asks before doing anything it isn't already allowed to.

Working with subagents

For larger tasks, the agent can spin up a focused subagent to handle a self-contained piece of work — for example, "investigate why this test fails" — and fold the result back into the main conversation. Each subagent gets its own recorded session, so you can inspect what it did.

→ Subagents & multi-agent

Long-running work in the background

Some commands take a while. gdc can move a long-running command to the background so the conversation keeps going, then deliver the result when it finishes. In the terminal app you can background a running command and review jobs with keyboard shortcuts.

→ Interactive terminal

Extending the tool set

Beyond the built-ins, you can give the agent more tools:

• MCP servers — connect external tools and data sources (issue trackers, databases, documentation services).

• Plugins — add self-contained tools packaged as portable modules.

• Skills — give the agent named instruction sets it can pull in on demand.

• Language servers — feed the agent editor-grade signals (definitions, references, diagnostics) about your code.

Memory tools also let the agent read and write durable notes — see Memory.

Staying in control

• Read-only tools run freely; anything that can change your machine is gated. See Permissions.
• You can restrict which tools are available at all using a profile — for example, a review profile with no ability to run commands.
• Every tool run is recorded in the session, so there's always an audit trail.