Once you let Claude do things instead of just say things, the next question is where those things actually happen. That answer, in your environment or on Anthropic's side, decides your control, your data boundary, and about half of any security review.
It is a simple split with real consequences, so it is worth getting straight.
A tool is an action, not an answer
On its own, a model produces text. A tool is what lets it act on the world instead: fetch a web page, query your CRM, edit a file, run a snippet of code. Tools are what turn Claude from something that answers into something that does, which is the whole basis of an agent.
There are two places a tool can run, and the difference is the useful part of this whole topic.
Client-side tools run where you are
A client-side tool executes in your environment. You define it and your side runs it: a function in your code, a file edit Claude Code performs on your machine, an API call to your CRM, or any tool you connected over MCP. The credentials and the data stay on your side; Claude decides what to call and reads back the result.
This is where most of a GTM stack lives. Your HubSpot, your Clay, your Slack, your files, all show up as client-side tools, usually through MCP.
- Runs in your environment; you (or Claude Code) execute it Client-side
- Runs on Anthropic's side; Claude calls it directly Server-side
Server-side tools run on Anthropic's side
Some tools run on Anthropic's infrastructure, and Claude calls them directly with nothing for you to wire up. You switch them on and they work.
- Web search: pull current information from the open web
- Code execution: run code in a sandbox to compute or transform data
- Computer use: operate a screen, click and type, for tasks with no API
- Web fetch: read a specific page you point it at
Why the difference matters
Three things ride on where a tool runs. None of them makes one kind better than the other; they make each right for a different job:
- Data boundary: client-side keeps your data and keys on your side, which is what a security review actually asks about.
- Setup: server-side tools need none; client-side tools you connect.
- Control: client-side means you decide exactly what the tool can touch; server-side means you accept Anthropic's implementation.
The line that matters more: read vs write
Whether a tool runs on your side or Anthropic's is worth knowing, but the line that should set your pulse is read versus write. A tool that only reads, a web search, a CRM lookup, a transcript pull, is almost impossible to get in trouble with. A tool that writes, changes a budget, emails a customer, updates a record, is where the value is and where an agent can do damage.
Grant read freely. Grant write on purpose, and put an approval gate in front of anything irreversible.
TipSort your tools into read and write before you sort them into client and server. The read ones you can turn on today; the write ones deserve a guardrail first.
Which tools a GTM automation actually uses
A real GTM workflow mixes both. Server-side web search does the live account research. Client-side tools over MCP do the work that touches your systems: read the CRM, score against your ICP, write the note, post to Slack. The agent orchestrates across them; you decide which ones get write access.
That is the shape behind the automations on this site: server-side reach for research, client-side reach for your stack, and a human on the writes that matter.
Frequently asked questions
What are Claude's tools?
The actions Claude can take beyond generating text: reading a web page, calling an API, editing a file, running code. Tools are what turn a model that answers into an agent that does, and they come in two kinds by where they run.
What is the difference between client-side and server-side tools?
Client-side tools run in your environment, executed by your code or by Claude Code, so data and credentials stay on your side (your CRM API, file edits, MCP servers). Server-side tools run on Anthropic's infrastructure and Claude calls them directly with no setup (web search, code execution, computer use).
What server-side tools does Claude have?
The common ones are web search (current information from the web), code execution (run code in a sandbox), computer use (operate a screen for tasks with no API), and web fetch (read a specific page). You enable them; there is nothing to wire.
How do I add my own tool?
As a client-side tool. The clean way is MCP: connect a server for the tool once and Claude can use it. In code, you define the tool (function calling) and your side executes it when Claude asks. Either way the tool runs in your environment.
Do server-side tools see my private data?
They act on what you pass them and run on Anthropic's side, so they do not reach into your systems unless you hand over the data. To keep private data and credentials in your boundary, use client-side tools (over MCP) for anything that touches your CRM, files, or internal systems.
Which tools should I use for a GTM workflow?
Both, by job. Use server-side web search for live account and competitor research, and client-side tools over MCP for anything that touches your stack, reading the CRM, scoring a lead, writing a note, posting to Slack. Keep a human on the writes that change a live system.
Sources & further reading
Claude ships fast. This page was last reviewed Aug 22, 2026; verify time-sensitive details against the official docs above before relying on them.