> ## Documentation Index
> Fetch the complete documentation index at: https://docs.neo.projectdiscovery.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Overview

> The capabilities Neo agents use to do real work, from pre-installed security binaries to tools you build yourself

Tools are how Neo agents get work done. The agent reasons about what to do and why, and tools handle the how: running a scanner, sending an HTTP request, driving a browser, searching code, or filing an issue. Each [agent](/platform/agents) is given a focused set of tools that match its job, so the Sandbox Agent can execute commands while the Research Agent can search the web.

Neo tools come from four sources. Understanding the difference helps you reason about what Neo can do and how to extend it.

<CardGroup cols={2}>
  <Card title="Pre-installed Tools" icon="screwdriver-wrench" href="/platform/tools/pre-installed-tools">
    The full security toolchain baked into every sandbox, invoked from the terminal
  </Card>

  <Card title="MCP Tools" icon="plug" href="/platform/tools/mcp-tools">
    External tools connected over the Model Context Protocol
  </Card>

  <Card title="Purpose-built Tools" icon="toolbox" href="/platform/tools/purpose-built-tools">
    Over 40 native tools designed by ProjectDiscovery for security work
  </Card>

  <Card title="Custom Tools" icon="wand-magic-sparkles" href="/platform/tools/custom-tools">
    Tools you write yourself to wrap internal APIs and team workflows
  </Card>
</CardGroup>

## How tools run safely

Every tool, regardless of where it comes from, runs through the same execution layer that keeps the agent durable. Tools have **timeouts** so a slow command cannot stall a task, and their output is **truncated** to fit the model's context while the full output still streams to you. Your [secrets](/platform/settings/secrets) are injected where a tool needs them and **redacted** from anything the model sees. Tools never crash the agent: a failure comes back as a structured error the agent can reason about and recover from. For sensitive actions, a tool can **pause and ask for your approval** before it proceeds.