DEV Community

Zac
Zac

Posted on

MCP servers with Claude Code: the ones worth adding and the ones to skip

#claudecode #mcp #productivity #programming

Model Context Protocol (MCP) lets you connect Claude Code to external tools — databases, APIs, services — that it can call during a session. The ecosystem is expanding fast. Not everything is worth adding.

Here's what's actually useful.

What MCP is (briefly)

MCP servers are local processes that Claude can call as tools. You configure them in .claude/settings.json and Claude uses them like any other tool — the difference is they talk to real systems.

{
  "mcpServers": {
    "postgres": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-postgres", "postgresql://localhost/mydb"]
    }
  }
}
Enter fullscreen mode Exit fullscreen mode

The ones that are genuinely useful

Database MCP servers (Postgres, SQLite, MySQL). Having Claude query your database directly changes how you work. Instead of pasting query results, it can read the schema, run queries, and understand your actual data. Very useful for debugging data issues and writing migrations.

Filesystem MCP. Extends what Claude can access beyond the current project. Useful when you're working across multiple repos or need to reference files outside the project root.

GitHub MCP. Read issues, PRs, and code directly. Good for "fix the bug described in issue #123" workflows where Claude needs to read the issue context.

Browser/search MCP. Lets Claude look up documentation, search for packages, or verify an API endpoint exists. Removes the back-and-forth of pasting docs.

When to add them

Add an MCP server when you notice yourself repeatedly copying and pasting information from an external system into the Claude session. That friction is what MCP eliminates.

Don't add MCP servers speculatively. Every server is another process running locally, another attack surface, another thing that can fail. Add them when the value is clear.

Security considerations

MCP servers can access real systems. A database MCP pointing at production can run queries against production. A filesystem MCP can read files anywhere on your machine that Claude can access.

Practical rules:

  • Point database MCPs at dev/staging, not production
  • Review what access each MCP server grants before adding it
  • Don't add community MCP servers from unknown sources to sessions with access to sensitive data

The MCP servers I actually use

  • Postgres MCP for all database work — schema exploration, query debugging, migration review
  • GitHub MCP for issue-driven development
  • Brave Search MCP when I need Claude to look something up mid-session

Everything else I've tried added complexity without enough value to justify it.

Getting started

If you haven't used MCP before, start with one server — the database one if you're doing any backend work. Get comfortable with it before adding more.

Full MCP setup guide: builtbyzac.com/agent-harness.html. MCP starter kit: builtbyzac.com/mcp-starter.html.

Top comments (0)