What are Claude's custom tools?

Custom tools are capabilities you add to Claude through external configuration. The three main types are MCP servers (plugins that connect Claude to databases, APIs, and services), API tool use (programmatic function calling for developers building AI-powered products), and Claude Projects with custom instructions (persistent workspaces with uploaded knowledge and behavioral rules).

Do I need custom tools or are built-in tools enough?

Built-in tools handle most general tasks — writing, analysis, code generation, research. You need custom tools when Claude must interact with your specific systems: reading your database, posting to your Slack, managing your files, or accessing proprietary data. Start with built-in tools and add custom tools when you hit their limits.

Are MCP servers the same as Claude plugins?

MCP servers are Claude's equivalent of plugins, but built on an open standard. Unlike proprietary plugin systems, MCP (Model Context Protocol) works across multiple AI tools — Claude Desktop, Claude Code, VS Code, and others. One MCP server works everywhere MCP is supported.

Can I use built-in and custom tools together?

Yes, and that is where the real power is. For example, you can use an MCP server to pull data from your database, then use Claude's built-in analysis tool to run Python on that data and generate visualizations — all in one conversation. The tools compose naturally.

Claude's Built-in Tools vs Custom Tools: What's the Difference?

Q: What are Claude's built-in tools?

Claude's built-in tools are native capabilities available in every conversation without any setup. They include artifacts (interactive previews of code, documents, and visualizations), analysis tool (Python code execution for data work), web search (real-time information retrieval), file handling (reading and processing uploaded documents), and advanced reasoning for complex problem-solving.

I talk to people every week who know Claude is powerful but cannot figure out where the built-in capabilities end and the custom extensions begin. The confusion is understandable. Anthropic has shipped a lot of functionality fast, and the line between what comes out of the box and what you configure yourself is not always obvious.

Here is the complete breakdown.

What Are Claude's Built-in Tools?

Built-in tools are capabilities that exist in every Claude conversation. You do not install them, configure them, or pay extra for them. They are part of the model.

Artifacts

When you ask Claude to write code, create a document, or build a visualization, it can render that output as an interactive artifact — a live preview that sits alongside the conversation. You can run HTML/CSS/JavaScript directly, view SVGs, and iterate on the output without copy-pasting anything. This is not a file export. It is a working preview inside the chat.

Analysis Tool (Code Execution)

Claude can write and execute Python code in a sandboxed environment. Upload a CSV and ask for trends. Request a statistical analysis. Generate matplotlib charts. The analysis tool handles data processing that would be unreliable with pure language generation. When numbers matter, Claude runs actual code instead of guessing.

Web Search

Claude can search the web in real time and cite sources. This matters because the model's training data has a cutoff. Need today's stock price, a recent news article, or the current documentation for a library? Web search fills the gap between what Claude knows and what exists right now.

File Handling

Upload PDFs, images, spreadsheets, code files, or plain text. Claude reads them, understands their structure, and works with their content directly. This is not an OCR hack — the model processes documents natively, handling everything from financial statements to architectural diagrams.

Vision and Reasoning

Claude can analyze images, interpret charts, read handwriting, and work through multi-step reasoning problems. These are not separate tools you enable. They are always on.

What Are Custom Tools?

Custom tools are capabilities you add to Claude by connecting it to external systems. They require setup, but they are what transform Claude from a smart conversational partner into an operational layer for your work.

MCP Servers

Model Context Protocol is the integration standard Anthropic built for connecting AI to external tools. An MCP server is a small program that exposes your systems to Claude — your database, your file system, your calendar, your CRM, whatever you need.

Think of MCP servers as USB ports. Each one gives Claude a new capability: read Slack messages, query PostgreSQL, manage GitHub repos, send emails. The protocol is standardized, so one MCP server works across Claude Desktop, Claude Code, and other compatible tools.

API Integrations (Tool Use)

If you are building a product with Claude inside it, the API's tool use feature lets you define custom functions that Claude can call. You describe the function (name, parameters, purpose), Claude decides when to call it, and your code executes the actual logic. This is how developers build AI agents, chatbots with real capabilities, and automated workflows.

Projects with Custom Instructions

Claude Projects let you create persistent workspaces with custom instructions and uploaded knowledge files. Every conversation in that project follows your rules — your brand voice, your formatting standards, your domain context. This is the simplest form of customization: no code, no servers, just configuration that shapes behavior.

Key Differences

| | Built-in Tools | Custom Tools | |---|---|---| | Setup required | None | Configuration or code | | Available to | All Claude users | Users who configure them | | Scope | General-purpose | Your specific systems | | Examples | Web search, code execution, file reading | MCP servers, API tool use, Projects | | Data access | Public web, uploaded files | Your databases, APIs, internal tools | | Technical skill needed | Zero | Low (Projects) to moderate (MCP/API) | | Maintenance | Anthropic handles it | You maintain your integrations |

The fundamental difference: built-in tools work with information Claude can already access. Custom tools give Claude access to information and systems it otherwise cannot reach.

When to Use Each

Use built-in tools when:

You need writing, analysis, or code generation from a conversation
Your data can be uploaded directly (files under the size limit)
You want real-time web information
The task is self-contained — no external system interaction needed

Use custom tools when:

Claude needs to read or write to your specific systems (databases, APIs, repos)
You want persistent behavioral customization across conversations
You are building a product with AI capabilities
Your workflow involves actions, not just answers (sending emails, creating tickets, deploying code)

Most people start with built-in tools and graduate to custom tools as their needs grow. That progression is natural.

How to Set Up Custom Tools

The fastest path to custom tools depends on your technical level:

No code: Create a Claude Project with custom instructions and knowledge files. Five minutes, zero technical skill.

Some technical comfort: Install pre-built MCP servers. The MCP ecosystem has servers for GitHub, Slack, PostgreSQL, Google Drive, and dozens of other services. Configuration is a JSON file.

Developer: Build your own MCP server or use the API with tool use. A minimal MCP server is about 15 lines of TypeScript. The API's tool use feature works in any language with HTTP support.

For a deeper walkthrough, see the complete Claude tools ecosystem guide.

The Real Power: Combining Both

The distinction between built-in and custom tools matters for understanding, but in practice the power is in composition.

A real workflow might look like this: an MCP server pulls your weekly sales data from PostgreSQL. Claude's built-in analysis tool runs Python to identify trends and generate charts. The artifacts system renders those charts as interactive visualizations. Another MCP server posts the summary to your team's Slack channel.

Four tools — two built-in, two custom — working together in a single conversation. No manual data exports. No switching between applications. No copy-pasting numbers into a spreadsheet.

This is what AI-native workflow design looks like. Not picking one tool. Composing all of them.

If you want to learn how to build these kinds of workflows from scratch, the Like One Academy covers everything from basic prompting through MCP server development and API integration — structured as hands-on courses, not lecture slides.