Which MCP Servers I Actually Use with Claude Code

Not another "top 10 MCP servers" listicle. This is my actual daily setup — 8 servers that survived months of real work. I'll show you which ones I use, why each one earned its place, and give you my exact configuration.

My Current Setup

Here's what claude mcp list returns on my machine right now:

chrome-control (Puppeteer)     ✓ Connected
context7                       ✓ Connected
perplexity-ask                 ✓ Connected
perplexity-local               ✓ Connected
firebase                       ✓ Connected
GA4 JobLander                  ✓ Connected
gsc                            ✓ Connected
dataforseo                     ✓ Connected

Eight servers. I started with twelve. Some didn't survive the cut. More on that later.

The Search Stack

Context7 — Documentation Lookup

This one changed how I work with unfamiliar libraries.

Claude's knowledge has a cutoff date. Ask it about Next.js 15 server actions, and it'll give you Next.js 13 patterns. Context7 fixes this by fetching up-to-date documentation directly into your conversation.

When I'm working with a library I haven't touched in months, I ask Claude to check Context7 first. It finds the current API, not what it remembers from training data.

claude mcp add context7 -- npx @upstash/context7-mcp

Perplexity — Web Search

I run two Perplexity MCPs. One for quick searches, one for deeper research.

The official perplexity-ask handles straightforward queries — "how do I X in Y" questions. When I need to understand a problem, compare solutions, or check what the community is saying, Perplexity searches the web and summarizes it.

Beats alt-tabbing to a browser. The AI finds what it needs without me copying and pasting context back and forth.

claude mcp add perplexity-ask -- npx server-perplexity-ask

The Automation Stack

Puppeteer (chrome-control)

Browser automation from inside Claude Code. I use it for two things: taking screenshots and running quick e2e tests.

When I finish a UI feature, I ask Claude to screenshot it. Visual verification without opening the browser myself. For testing, it can click through flows, fill forms, check if things break.

The official Anthropic Puppeteer server works well:

claude mcp add chrome-control -- npx @modelcontextprotocol/server-puppeteer

The Analytics Stack

This is where my setup differs from most guides. I don't just code — I care about whether people find and use what I build.

Google Search Console

My site's SEO data, accessible from Claude Code.

I check impressions, clicks, positions for my articles. When writing new content, I verify if my target keywords are getting traction. Claude can pull this data, analyze trends, and suggest what's working.

claude mcp add gsc -- npx -y mcp-server-gsc

You'll need to authenticate with your Google account the first time.

Google Analytics

Traffic data. I use it less frequently than GSC, but it's useful for understanding user behavior beyond just search.

claude mcp add ga4 -- npx -y mcp-server-google-analytics

DataForSEO

This one's specialized. Keyword research, SERP analysis, competitor data.

I used it to research this very article. Checked keyword volumes, difficulty scores, analyzed what's ranking. Claude calls DataForSEO APIs, processes the data, and helps me decide which topics to target.

Not free — requires a DataForSEO account. But if you're serious about SEO, it pays for itself.

claude mcp add dataforseo --env DATAFORSEO_USERNAME=your_username --env DATAFORSEO_PASSWORD=your_password -- npx -y dataforseo-mcp-server

The Backend Stack

Firebase

I use Firebase for some projects. The official MCP server lets Claude interact with Firestore, check deployment status, manage hosting.

claude mcp add firebase -- npx -y firebase-tools@latest experimental:mcp

Requires Firebase CLI setup and project authentication.

What I Removed

Started with more servers. Trimmed down to what I actually use.

GitHub MCP: Seemed useful. In practice, Claude Code already handles git operations natively. The MCP added complexity without adding value.

Slack MCP: Tried it for a week. The context switching between code and messaging in the same session was more confusing than helpful. I keep Slack separate.

Linear/Jira MCPs: Good idea, poor execution. Reading tickets is useful, but I found myself just opening the app anyway for context.

The lesson: start minimal. Add servers when you feel the pain, not because they sound cool.

Configuration Tips

Keep your MCP setup project-specific when possible. Global config bloats your Claude sessions with servers you don't need for every task.

Check server health regularly. Run claude mcp list and make sure everything shows "Connected". Dead connections slow down Claude as it waits for timeouts.

Order matters less than you'd think. Claude figures out which server to use based on the task.

Getting Started

If you're new to MCP, don't install everything at once.

Start with Context7 and Perplexity. They're universally useful. Context7 for documentation, Perplexity for research. That covers most information-gathering needs.

Add Puppeteer when you need visual verification or testing.

Add analytics servers when you care about measuring what you build.

The MCP ecosystem is growing fast. New servers appear weekly. Most won't survive. Wait for the ones that prove themselves in real workflows.

My setup works for my projects — a mix of frontend, SEO, and analytics work. Yours will differ. The principle stays the same: install what you use, remove what you don't.

MCP servers extend Claude's reach into your actual tools. The right setup makes Claude feel less like a chatbot and more like a teammate who can access the same systems you do.

What servers are you using?