Claude Code MCP: Give Your AI Access to Any Tool

What is /mcp

When using Claude Code, you might run into situations like these:

You want Claude Code to query your database directly, but it can’t access it
You want it to manage GitHub issues, but it’s limited to the gh CLI
You want it to read docs from Confluence or Notion, but it can’t reach them

Claude Code has natural boundaries — it can read and write local files and run shell commands, but when it comes to external services and data sources, it’s out of reach.

That’s exactly what the /mcp command is for.

/mcp is a slash command in Claude Code for managing MCP (Model Context Protocol) servers. MCP servers act like “plugins” that let Claude Code connect to external tools and data sources, gaining capabilities it doesn’t have natively.

Understanding MCP First

MCP stands for Model Context Protocol, an open standard created by Anthropic.

Think of it as a USB port for AI — a universal protocol that lets AI models connect to all kinds of external tools and data sources. With MCP, Claude Code is no longer limited to local files and shell commands — it can talk to databases, APIs, cloud services, and more.

An MCP server is essentially a small program that:

Implements the MCP protocol
Provides a set of specific tools (e.g., “query database”, “create GitHub issue”)
Claude Code calls these tools through the protocol, just like its built-in tools

How to Use It

View Configured MCP Servers

In Claude Code’s interactive mode, type:

/mcp

This lists all currently configured MCP servers and their status.

Add an MCP Server (CLI)

Use the claude mcp add command in your terminal:

claude mcp add <name> <command> [args...]

For example, adding a filesystem MCP server:

claude mcp add filesystem npx -y @modelcontextprotocol/server-filesystem /path/to/dir

Add with JSON (More Flexible)

For complex configurations like environment variables, use add-json:

claude mcp add-json github-server '{
  "command": "npx",
  "args": ["-y", "@modelcontextprotocol/server-github"],
  "env": {
    "GITHUB_TOKEN": "your-token-here"
  }
}'

Remove an MCP Server

claude mcp remove <name>

List All Servers

claude mcp list

Configuration Scope

MCP server configurations can live in different locations, each with a different scope:

Scope	Storage Location	Affects	Flag
project	`.mcp.json` in project root	Current project only, can be committed to git	`--scope project`
user	`~/.claude/.mcp.json`	All projects	`--scope user`

Project-Level Config (Recommended for Teams)

claude mcp add filesystem npx -y @modelcontextprotocol/server-filesystem ./data --scope project

This writes to .mcp.json in the project root. Commit it to git and every team member gets the same MCP servers automatically.

User-Level Config (Personal Use)

claude mcp add my-db npx -y @modelcontextprotocol/server-postgres postgres://localhost/mydb --scope user

This writes to ~/.claude/.mcp.json, applies only to you, and works across all projects.

Useful MCP Servers

The MCP ecosystem is growing rapidly. Here are some popular ones:

MCP Server	Purpose	Package
Filesystem	Controlled file read/write	`@modelcontextprotocol/server-filesystem`
GitHub	Manage repos, issues, PRs	`@modelcontextprotocol/server-github`
PostgreSQL	Query PostgreSQL databases	`@modelcontextprotocol/server-postgres`
SQLite	Query SQLite databases	`@modelcontextprotocol/server-sqlite`
Brave Search	Web search	`@modelcontextprotocol/server-brave-search`
Puppeteer	Browser automation	`@modelcontextprotocol/server-puppeteer`
Memory	Knowledge graph-based persistent memory	`@modelcontextprotocol/server-memory`
Slack	Read and send Slack messages	`@modelcontextprotocol/server-slack`

For more, search awesome-mcp-servers on GitHub — the community maintains several curated lists.

Real-World Use Cases

Use Case 1: Connect a Database for Data Analysis

After adding a PostgreSQL MCP server, you can simply tell Claude Code:

Show me the number of users who registered in the past 7 days, grouped by day

Claude Code queries the database directly through the MCP server and returns the results. No need to write SQL yourself or copy-paste query results back and forth.

Use Case 2: Let Claude Code Manage GitHub

With the GitHub MCP server added, Claude Code gains more capabilities:

Create and close issues
View PR details and comments
Search code across repositories
Read files from other repos

Use Case 3: Connect a Knowledge Base

If your team docs live in Notion or Confluence, the corresponding MCP server lets Claude Code read them directly — no more manual copy-pasting.

Practical Tips

Tip 1: Commit Project Config to Git

Commit .mcp.json to your repo so team members get it on clone. But be careful — never put API keys or passwords in .mcp.json. Pass sensitive information through environment variables, or keep them in user-level config.

Tip 2: Use Environment Variables for Secrets

claude mcp add-json my-api '{
  "command": "npx",
  "args": ["-y", "some-mcp-server"],
  "env": {
    "API_KEY": "${MY_API_KEY}"
  }
}'

Keep actual secrets in .env or your shell profile. Reference only variable names in .mcp.json.

Tip 3: Add Only What You Need

Each MCP server takes up Claude Code’s context window (because it needs to describe available tools). Too many unused servers waste tokens. Only add what your current project actually needs.

Tip 4: Use /mcp to Troubleshoot

If an MCP server isn’t working, type /mcp in interactive mode to check its status. It shows whether each server is connected and what tools it provides. Failed connections will display clear error messages.

What .mcp.json Looks Like

A typical project-level .mcp.json file:

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "./docs"]
    },
    "database": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-postgres"],
      "env": {
        "DATABASE_URL": "${DATABASE_URL}"
      }
    }
  }
}

The structure is straightforward: under the mcpServers object, each key is a server name and each value is its launch config — command, arguments, and environment variables.

/mcp + CLI vs Manual Config

	/mcp + CLI Commands	Manually Edit .mcp.json
Ease of use	CLI interaction, less error-prone	Hand-write JSON, possible format errors
Speed	One command and done	Find file, edit, save
Flexibility	Covers common scenarios	Full control over config
Best for	Day-to-day add/remove	Bulk config, complex setups

Both approaches write to the same config file, so pick whichever suits the situation. CLI commands for daily use, direct file editing for bulk or fine-tuned configuration.

Final Thoughts

/mcp is one of the most extensible commands in Claude Code. The problem it solves is fundamental: it lets Claude Code reach beyond local files and terminal commands to connect with your entire development ecosystem.

Databases, GitHub, documentation systems, cloud services — as long as there’s a matching MCP server, Claude Code can interact with it directly. This upgrades an AI coding assistant from “can only see code” to “can touch every part of your workflow.”

Next time you catch yourself thinking “if only Claude Code could query my database directly,” remember MCP — there’s probably a server for that already.