Claude Code MCP: Connect Your Tools

TL;DR

• Claude Code MCP lets your AI connect directly to external tools and data sources.
• You add MCP servers with a single JSON config entry in your project.
• Useful starting servers include GitHub, Notion, filesystem, and Postgres.
• Once connected, Claude can read, write, and act without you copying and pasting.
• Custom MCP servers let you connect anything your business runs on.

---

Claude Code is already powerful on its own. But Claude Code MCP is what makes it genuinely useful in a real work environment.

Without MCP, Claude works with whatever you paste into the conversation. That is fine for one-off tasks. It breaks down fast once you are working across GitHub repos, Notion docs, databases, and live APIs every single day.

MCP fixes that. It gives Claude a direct line to your actual tools.

---

What does MCP add to Claude Code?

MCP stands for Model Context Protocol. Think of it as a plugin system for AI.

Each MCP server exposes a set of tools. Claude can call those tools during a session. It can read files from your filesystem, query a database, pull issues from GitHub, or check a Notion page, all without you doing any copy-paste work in between.

Before MCP, you were the bridge between Claude and your tools. You grabbed the data, pasted it in, waited for an answer, then went and applied it yourself. With MCP, Claude is the bridge. You just tell it what you need done.

That shift sounds small. In practice it changes how you work entirely. Tasks that used to take ten minutes of back-and-forth now happen in a single prompt.

MCP turns Claude Code from a smart text box into an AI that actually touches your work.

---

How do you add an MCP server?

Adding an MCP server to Claude Code takes about two minutes. Claude Code reads MCP configuration from a JSON file. You can set this up at the project level (inside your repo) or globally for your whole machine.

Step 1: find or create your config file. For project-level config, create a file named .mcp.json in your project root. For global config, it lives in your home .claude folder.

Step 2: add your server entry. Inside the file there is a top-level key called mcpServers. Each server you add gets its own named entry. Each entry needs a command to run the server and any arguments or environment variables it needs. For the official filesystem server, the command is npx, and the arguments point at the filesystem server package plus the path to the folder you want Claude to access.

Step 3: check it works. Restart Claude Code and start a new session. Ask Claude to list files in your project directory. If it responds with real file names rather than asking you to paste them, your server is running.

Step 4: add more servers. You can stack as many servers as you need in the same config file. Each one gets its own key inside mcpServers. Some servers need API keys or environment variables. Pass those in an env block inside the server config. Keep secrets out of your repo by referencing environment variables rather than hardcoding values.

The config file is plain JSON. No build step, no special tooling. Just add a block and restart.

---

Which MCP servers should you start with?

There are hundreds of community-built MCP servers now. Most people do not need more than four or five to start. Here are the ones worth setting up first.

Filesystem. The official filesystem server lets Claude read and write files in directories you specify. This is the baseline. Most Claude Code workflows need it. You control which paths it can access. Keep it scoped to your working directory.

GitHub. The GitHub MCP server connects Claude to your repos. It can read issues, create pull requests, search code, and post comments. Claude can look up the issue context before writing code, then open a PR when it is done.

Notion. The Notion MCP server gives Claude read and write access to your workspace. It can pull page content, update databases, and create new pages. If you run processes in Notion, this one saves a lot of time.

Postgres or SQLite. Database MCP servers let Claude query your data directly. It can write SQL, run it, and interpret the results in one go. Useful for pulling reports or any data work where you would normally export a CSV.

Fetch. The fetch server lets Claude make HTTP requests to any URL. It can pull web pages, call REST APIs, and check live data. Good for anything involving external APIs that do not have a dedicated MCP server yet.

For links to the official servers and the community registry, see the Model Context Protocol guide and the deeper breakdown in what is an MCP server.

Start with filesystem and one domain-specific server. Add more once you know what slows you down.

---

What can you actually do once servers are connected?

Here are a few real examples from daily use.

Claude can open a GitHub issue, read the full thread, write the fix, test it, and open a PR. You review the PR. That is the whole loop.

Claude can pull a Notion brief, write a blog post that matches the spec, and save the draft back to Notion. One prompt.

Claude can query your database for last month's numbers, write the analysis, and drop it into a report file. No spreadsheet work.

Claude can check your filesystem for outdated config files, compare them to your current setup, and flag anything that needs updating.

None of that requires you to shuttle data between tools. Claude handles the movement. That is why MCP matters. It is not about making Claude smarter. It is about removing the friction between Claude and the work. For a deeper look at how Claude Code handles multi-step tasks like these, read the Claude Code subagents guide or the complete Claude Code guide.

---

What about custom MCP servers?

The official and community servers cover a lot of ground. But most businesses run tools that are not covered.

Custom CRMs, internal databases, proprietary APIs, legacy systems. If your data lives somewhere that does not have an MCP server yet, you can build one.

A custom MCP server is a small service that wraps your tool in the MCP protocol. Once it is running, Claude treats it like any other server. It can read data, write records, trigger actions, whatever the server exposes.

This is where the real value is. Once Claude can touch your actual business data, the whole category of AI assistant changes. It stops being a writing tool and starts being an operator. At Devwiz, we build custom MCP servers for businesses that want to connect Claude to their specific stack.

---

FAQ

Can I use multiple MCP servers at the same time?
Yes. You can add as many servers as you need to your config file. Claude has access to all of them in the same session. Just keep the config tidy so it is easy to manage.

Do MCP servers work with Claude Code on any project?
Project-level config applies to that project only. Global config applies everywhere. Use project-level for servers specific to one codebase, global for tools you use across everything.

Is it safe to give Claude write access through MCP?
Yes, with the right setup. Scope filesystem access to specific directories. Use read-only database users where you do not need writes. Review what each server can do before adding it. Claude will tell you what it is about to do before it does it.

What if there is no MCP server for the tool I use?
You have two options. Use the fetch server to call the tool's REST API directly. Or build a custom MCP server that wraps your tool. Custom servers are not as hard as they sound, especially for simple read and write operations. Devwiz can help if you want someone to build it for you.

---

Ready to connect Claude Code to your business tools? If you want a custom MCP setup built for your stack, book a call and let us sort it out.