Model Context Protocol

The MCP Servers Guide
Nobody Wrote

Model Context Protocol (MCP) turns Claude Code from a code editor into a connected agent. Here's every server worth installing, how they work, and how to set them up in 5 minutes.

Read the Guide → Full Claude Code Guide
In this guide
Foundations

What is Model Context Protocol?

MCP (Model Context Protocol) is an open standard created by Anthropic. It defines how AI assistants connect to external tools, databases, APIs, and data sources using a standardized protocol — the same way HTTP standardized how browsers talk to servers.

Before MCP, every AI tool had to build its own integrations from scratch. MCP makes those integrations universal: one server, any client. Claude Code, Claude Desktop, Cursor, and other MCP-compatible clients can all use the same MCP servers.

The key insight

Claude Code uses MCP to connect to databases, run browser automation, read files from remote servers, and call external APIs — all directly inside your conversation. You describe what you want; Claude uses the right tool to do it.

MCP servers expose three types of capabilities to Claude:

⚙️

Tools

Functions Claude can call — run a query, create a file, send a message, trigger an API.

📂

Resources

External data Claude can read — files, database records, API responses, knowledge bases.

📋

Prompts

Reusable templates that surface as slash commands in Claude's interface.

Most users only ever interact with Tools — Claude calls a function, gets a result, continues the conversation. Resources and Prompts unlock more advanced patterns as you go deeper.

Here's how an MCP server entry looks in your config file:

~/.claude/settings.json — minimal MCP entry
// Add this to mcpServers in your settings.json { "mcpServers": { "filesystem": { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/you/projects"] } } }
Setup

How MCP Works with Claude Code

Claude Code reads MCP configuration from .claude/settings.json in your project directory, or from your global Claude Desktop config. MCP servers launch as subprocesses — Claude communicates with them over stdio.

1

Install the MCP server

Most servers are npm packages. Run the install command or just use npx -y to auto-install on first use — no global install needed.

2

Add the config entry

Add the server to mcpServers in your .claude/settings.json (project-level) or ~/Library/Application Support/Claude/claude_desktop_config.json (Claude Desktop).

3

Claude auto-discovers the tools

On next startup, Claude Code launches the MCP server subprocess and queries its available tools, resources, and prompts. No restart required after that — tools appear in the conversation immediately.

4

Use tools naturally in conversation

Ask Claude to query your database, browse a URL, read a Google Doc, or post to Slack. Claude selects the right MCP tool automatically — you don't need to know tool names or call syntax.

Config file locations

Claude Code (project): .claude/settings.json in your repo — applies to that project only.

Claude Code (global): ~/.claude/settings.json — applies to all Claude Code sessions.

Claude Desktop: ~/Library/Application Support/Claude/claude_desktop_config.json (macOS).

Both config locations use the same mcpServers key and identical JSON structure. Configure once, works in both clients.

Installation Guide

Top 8 MCP Servers to Install Today

These are the highest-leverage MCP servers — maintained by Anthropic or widely adopted by the community, with stable APIs and real daily utility.

📁

Filesystem MCP

Official

Read and write any file on your system within a directory scope you define. The foundational MCP server — gives Claude real file access beyond the current repo.

Best for: Code editing across multiple repos, document management, project scaffolding, reading logs and configs.

npx @modelcontextprotocol/server-filesystem /path/to/dir
🎭

Playwright MCP

Community

Full browser automation via Playwright. Claude can navigate URLs, click buttons, fill forms, take screenshots, and extract page content — without you writing a line of Playwright code.

Best for: Web scraping, end-to-end testing, form automation, visual QA, extracting data from web apps.

npx @executeautomation/playwright-mcp-server
🐙

GitHub MCP

Official

Full GitHub API access. Claude can create PRs, comment on issues, read repository contents, manage branches, trigger workflows, and review code — all from conversation.

Best for: PR reviews, issue triage, automated changelogs, repo analytics, cross-repo code searches.

npx @modelcontextprotocol/server-github
🐘

PostgreSQL MCP

Official

Direct read-only database access. Claude can explore your schema, run SQL queries, analyze data distributions, and debug query performance — using natural language, not SQL.

Best for: Data analysis, schema exploration, query debugging, building reports from live data.

npx @modelcontextprotocol/server-postgres postgresql://localhost/mydb
🔍

Brave Search MCP

Official

Real-time web search via the Brave Search API. Breaks Claude's knowledge cutoff — gives access to current events, recent docs, live prices, and anything that happened after training.

Best for: Research tasks, fact-checking against current sources, competitive analysis, documentation lookups.

npx @modelcontextprotocol/server-brave-search
💬

Slack MCP

Official

Read and post Slack messages. Claude can summarize channel activity, draft and send messages, search conversation history, and monitor threads — without leaving your coding session.

Best for: Team notifications, daily standup summaries, incident response, monitoring important channels.

npx @modelcontextprotocol/server-slack
📊

Google Drive MCP

Official

Access Google Docs, Sheets, and Drive files. Claude can read documents, extract spreadsheet data, search your Drive, and use document content as context for any task.

Best for: Document editing, spreadsheet analysis, PRD review, syncing project docs with code changes.

npx @modelcontextprotocol/server-gdrive
🧠

Memory MCP

Official

Persistent knowledge graph that survives across sessions. Claude can store facts, relationships, and project state — and recall them in any future conversation. Solves the context window problem for long-running projects.

Best for: Project memory, entity tracking, accumulating knowledge about a codebase over weeks of sessions.

npx @modelcontextprotocol/server-memory
Configuration

Full Configuration Example

Here's a complete .claude/settings.json with three MCP servers configured and ready to use. Drop this in your project root (or ~/.claude/ for global use).

.claude/settings.json — production-ready multi-server config
{ "mcpServers": { // Filesystem: read/write files under your project root "filesystem": { "command": "npx", "args": [ "-y", "@modelcontextprotocol/server-filesystem", "/Users/you/projects" ] }, // GitHub: full API access — PRs, issues, repos "github": { "command": "npx", "args": [ "-y", "@modelcontextprotocol/server-github" ], "env": { "GITHUB_PERSONAL_ACCESS_TOKEN": "ghp_your_token_here" } }, // Brave Search: real-time web search (requires free API key) "brave-search": { "command": "npx", "args": [ "-y", "@modelcontextprotocol/server-brave-search" ], "env": { "BRAVE_API_KEY": "BSA_your_key_here" } } } }
API keys in env blocks

Use the "env" block to pass API keys — never hardcode them in the args array. Claude Code passes these as environment variables to the MCP subprocess. For GitHub, create a Personal Access Token with repo scope. For Brave Search, register at api.search.brave.com — free tier is sufficient for most uses.

The same structure works for Claude Desktop. Copy the mcpServers block into ~/Library/Application Support/Claude/claude_desktop_config.json with no other changes.

Discovery

Where to Find More MCP Servers

The MCP ecosystem is growing rapidly. Hundreds of community-built servers exist beyond Anthropic's official list. Here's where to look:

Anthropic Official List

Anthropic's curated repository of first-party and verified community MCP servers. Start here — everything is tested and documented.

github.com/modelcontextprotocol/servers →

Smithery

The MCP server marketplace. Search by category, see install counts, read reviews. Easiest way to discover community servers for specific tools.

smithery.ai →

GitHub Tag Search

Search GitHub for the mcp-server topic. Finds newly published servers before they appear in marketplaces.

github.com/topics/mcp-server →

npm Keyword Search

Search npm for packages tagged with the mcp keyword. Most Node.js-based MCP servers publish to npm.

npmjs.com/search?q=keywords:mcp →
Security note

MCP servers run as subprocesses with the permissions you grant them. Only install servers from sources you trust. Review what filesystem paths and API scopes you're granting before adding any server to a production environment. The official Anthropic servers are audited — community servers vary.

Practical Patterns

MCP Use Cases by Role

The right MCP stack depends on what you're building. Here are four high-leverage combinations that turn Claude Code into a domain-specific agent.

Developer

Full-Stack Agent

filesystem github postgres

Claude reads code across the entire repo, creates PRs with real diff context, queries the production database to understand data shape, and ships features end-to-end without context switching.

Researcher

Research Engine

brave-search memory filesystem

Claude searches the web for current sources, remembers findings across sessions, and writes structured research reports to disk. Runs multi-step research without forgetting earlier steps.

Founder / PM

Cross-Team Intelligence

slack github google-drive

Claude monitors Slack for blockers, cross-references GitHub activity, summarizes what the team shipped this week, and writes the weekly update directly into your Google Doc.

Data Analyst

Analytical Workflow

postgres filesystem memory

Claude queries live data, builds analysis files, tracks metric definitions across sessions, and produces narrative reports — no SQL required, no context loss between days of analysis.

Maximum Leverage

MCP gives Claude tools.
Brainfile gives Claude knowledge.

MCP defines what Claude can do — call APIs, query databases, browse the web. Brainfile defines what Claude knows — your codebase conventions, domain vocabulary, product decisions, team patterns.

MCP
What Claude can DO
Brainfile
What Claude KNOWS
Together
Maximum leverage

Brainfile is a library of production-ready CLAUDE.md templates, .cursorrules files, and system prompt structures — built for every major development context. Drop them into your project alongside your MCP config and Claude immediately understands your domain at expert level.

$149/mo
Get Brainfile Templates — $149/mo → See what's included