What Are Claude Code Agents?
A Claude Code agent is a specialized sub-agent that Claude can spin up to handle a discrete task autonomously. When you trigger an agent, it receives a goal, formulates a plan, and executes a series of tool calls — file reads, code edits, shell commands, web searches — until the task is complete. Agents are composable: one agent can delegate work to another, enabling complex pipelines that would be impractical to manage manually.
How Agents Work
When you invoke a Claude Code agent, you hand it a goal rather than a series of instructions. The agent reads the relevant parts of your codebase, decides what actions are needed, calls the appropriate tools — file edits, shell commands, web searches, MCP servers — and iterates until the goal is achieved.
Agents run as sub-processes within Claude Code's context system. They have access to the same tools as the main Claude instance but operate with a scoped objective, which keeps them focused and prevents them from making unintended changes outside their assigned task.
Because agents can call other agents, you can build pipelines: a planning agent might decompose a feature request and dispatch separate agents to write code, generate tests, and update documentation — all in parallel.
Types of Agents
Agents are built for tasks where multiple sequential or parallel steps are needed. Common categories include:
Code Generation Agents
Scaffold entire features, modules, or components from a description. These agents read your existing patterns and generate code that fits your codebase conventions.
Research & Analysis Agents
Search the web, read documentation, and synthesize findings into actionable summaries. Useful for dependency evaluation, API exploration, and competitive analysis.
Testing Agents
Write, run, and iterate on tests autonomously. These agents analyze code coverage, identify untested paths, generate test cases, and fix failing tests without manual intervention.
Refactoring Agents
Migrate codebases to new patterns, update deprecated APIs, and apply consistent style changes across large file sets — tasks too tedious to do manually.
DevOps & Deployment Agents
Manage CI/CD pipelines, provision infrastructure, monitor deployments, and respond to incidents by running sequences of shell commands and API calls.
Agents vs. Skills vs. Plugins vs. MCP Servers
Skills
Teach Claude how to handle a single, well-defined task through structured instructions. Best for repeatable, single-step interactions.
Plugins
Bundle multiple skills, configurations, and MCP servers into a single installable package. Best for setting up a complete workflow for a framework or language.
MCP Servers
Connect Claude to external data sources and tools — databases, APIs, file systems — through a standardized protocol. Best for real-time data access.
Agents
Autonomous sub-processes that plan and execute multi-step tasks independently. Best when the work requires multiple decisions, tool calls, or parallelism.
How to Use an Agent
Most agents are invoked through Claude Code's agent system. Install an agent, then reference it in your Claude Code session:
claude agent run <agent-name> --goal "your task description" Each agent's detail page on Claude Bazaar includes its install command, supported goals, and configuration options. Agents can also be embedded in CLAUDE.md files to run automatically during specific workflows.
Finding the Right Agent
With 0 agents available, browse the full directory on our Agents page. Filter by task type or technology to find agents that fit your workflow. For lighter-weight automation, also consider skills — they're faster to install and suitable for single-step tasks.