Effloow / Articles / Top 15 MCP Servers Every Developer Should Install in 2026

Top 15 MCP Servers Every Developer Should Install in 2026

The best MCP servers for developers in 2026 — curated picks with real claude mcp add install commands, use cases, and caveats. Browser automation, databases, search, workflow, and more.

· Effloow Content Factory
#mcp #claude-code #model-context-protocol #developer-tools #ai-agents #browser-automation #database #productivity

Top 15 MCP Servers Every Developer Should Install in 2026

There are over 10,000 MCP servers listed across directories like mcpmarket.com, mcpservers.org, and GitHub. Most of them are weekend projects that break the first time you try them. A handful are production-grade tools that will fundamentally change how you work with AI coding assistants.

This guide is not a directory listing. We tested these servers in our daily workflow at Effloow, where we run a fully AI-powered company with 14 agents. Every pick includes a real claude mcp add install command, a concrete use case, and honest notes about what does not work well. If a server is deprecated or has significant limitations, we say so.

What Is MCP and Why It Matters Now

The Model Context Protocol (MCP) is an open standard created by Anthropic that lets AI assistants connect to external tools and data sources through a unified interface. Think of it as USB-C for AI — one protocol, any tool.

What changed in 2026 is that MCP stopped being an Anthropic-only thing. In December 2025, Anthropic donated MCP to the Agentic AI Foundation (AAIF), a directed fund under the Linux Foundation co-founded by Anthropic, Block, and OpenAI. Since then, adoption has exploded:

  • OpenAI adopted MCP across the Agents SDK, Responses API, and ChatGPT desktop (source)
  • Google DeepMind confirmed MCP support in Gemini models (source)
  • Amazon AWS implemented MCP for AI agent integration with AWS services (source)
  • The MCP SDK has crossed 97 million cumulative downloads (Python + TypeScript combined) (source)
  • Pinterest deployed MCP in production for their engineering workflows, as covered by InfoQ in April 2026

This is not a niche protocol anymore. Every major AI platform supports it, and the ecosystem is growing faster than any developer tool standard since Docker.

How to Install MCP Servers in Claude Code

Every server in this guide uses the claude mcp add command. The basic syntax is:

# For local stdio servers (most common)
claude mcp add <name> -- npx -y @package/server

# For remote HTTP servers
claude mcp add --transport http <name> <url>

# With environment variables
claude mcp add <name> -e API_KEY=your-key -- npx -y @package/server

# Scoped to all projects (user-wide)
claude mcp add --scope user <name> -- npx -y @package/server

After adding a server, verify it with /mcp inside Claude Code. To manage your servers: claude mcp list, claude mcp get <name>, or claude mcp remove <name>.

For a deeper dive into MCP configuration within your project, see our guide on CLAUDE.md best practices and project setup, which covers how MCP servers fit into your overall project configuration.

Now let us get into the picks.

Browser Automation

1. Playwright MCP

What it does: Browser automation from Microsoft — navigate pages, click elements, fill forms, take screenshots, and run end-to-end tests, all driven by your AI assistant through structured accessibility snapshots.

Install command:

claude mcp add playwright -- npx @playwright/mcp@latest

Real use case: You are building a checkout flow and want Claude to verify it works. Instead of describing the page, Claude opens a browser, walks through the flow, takes screenshots at each step, and reports what broke. No Selenium scripts, no manual testing.

Caveats: Requires Node.js 18+. The old package name @modelcontextprotocol/server-playwright is deprecated — always use @playwright/mcp from Microsoft. Headless mode is the default; pass --headed if you need to watch the browser.

Why this over Selenium or Cypress? Playwright MCP lets your AI assistant drive the browser conversationally. You say "test the login flow" instead of writing test scripts. For developers who use Claude Code's advanced workflow features, this pairs naturally with subagents that can run browser tests in parallel.

GitHub: microsoft/playwright-mcp

2. Puppeteer MCP

What it does: Browser automation using Google's Puppeteer — screenshot capture, JavaScript execution, page navigation, and DOM interaction.

Install command:

claude mcp add puppeteer -- npx -y @modelcontextprotocol/server-puppeteer

Real use case: Quick screenshot generation for visual QA. Point Claude at a staging URL, ask it to capture screenshots at different viewport sizes, and compare them against the production version.

Caveats: This server is deprecated. The official @modelcontextprotocol/server-puppeteer package is no longer actively maintained, and the MCP team recommends migrating to Playwright MCP. We include it here because many existing tutorials still reference it, and some teams have Puppeteer-specific workflows. For new projects, start with Playwright.

GitHub: modelcontextprotocol/servers (archived)

Code & DevOps

3. GitHub MCP

What it does: GitHub's official MCP server connects your AI assistant to the GitHub API — manage repositories, create and review pull requests, search code, manage issues, and trigger workflows through natural language.

Install command:

claude mcp add-json github '{"type":"http","url":"https://api.githubcopilot.com/mcp/","headers":{"Authorization":"Bearer YOUR_GITHUB_PAT"}}'

Replace YOUR_GITHUB_PAT with a GitHub Personal Access Token that has the repo, read:org, and workflow scopes.

Real use case: During a code review, ask Claude to "check if there are any open PRs that touch the authentication module" or "create an issue for the bug we just found in the payment flow." No context switching to the browser.

Caveats: The PAT needs appropriate scopes for each operation. Read-only operations need less permission than creating PRs or triggering workflows. Start with minimal scopes and add as needed.

GitHub: github/github-mcp-server

4. Git MCP

What it does: Local Git repository operations — commit history, diffs, branch management, and repository analysis without leaving your AI conversation.

Install command:

claude mcp add git --scope user -- uvx mcp-server-git

Note: This uses uvx (Python's package runner), not npx. You need uv installed.

Real use case: Ask Claude "what changed in the last 5 commits on the feature branch" or "show me all files modified by the auth refactor." The Git MCP server gives Claude direct read access to your repository history without running shell commands.

Caveats: Read-only by default, which is the right design choice — you do not want your AI assistant force-pushing to main. For write operations (commits, pushes), Claude Code's built-in Git integration is more appropriate.

GitHub: modelcontextprotocol/servers/src/git

Filesystem

5. Filesystem MCP

What it does: Gives Claude controlled access to read, write, search, and manage files within directories you explicitly allow. Directory listing, file creation, moving files, and content search.

Install command:

claude mcp add filesystem -- npx -y @modelcontextprotocol/server-filesystem $(pwd)

The path argument ($(pwd) or any absolute path) defines which directories the server can access. You can pass multiple paths to grant access to several directories.

Real use case: Working on a monorepo where you need Claude to search across multiple packages, find configuration files, or reorganize project structure. The filesystem server provides structured file operations with proper error handling instead of raw shell commands.

Caveats: Only the directories you explicitly pass as arguments are accessible — this is a security feature, not a limitation. Claude Code already has file access built in, so this server is most useful when you need to expose filesystem operations to other MCP clients or want finer-grained access control.

GitHub: modelcontextprotocol/servers/src/filesystem

Databases

6. Supabase MCP

What it does: Connects Claude to your Supabase projects — query your Postgres database, manage edge functions, inspect table schemas, and interact with your backend through natural language.

Install command:

claude mcp add --transport http supabase https://mcp.supabase.com/mcp

After adding, run /mcp in Claude Code and follow the OAuth flow to authenticate. No personal access token needed — Supabase handles auth automatically.

Real use case: You are debugging a data issue and need to understand your schema. Ask Claude "show me all tables related to user subscriptions" or "what does the billing_events table look like." Claude can query your database, explain relationships, and even help write migrations.

Caveats: The OAuth flow requires a browser redirect, so it works best in environments where you can open a browser. For CI/CD or headless environments, you may need to use a service token approach instead.

GitHub: supabase/mcp

7. PostgreSQL MCP

What it does: Direct read-only access to any PostgreSQL database. Exposes your schema as resources and supports SQL query execution.

Install command:

claude mcp add postgres -- npx -y @modelcontextprotocol/server-postgres "postgresql://user:password@localhost:5432/mydb"

Replace the connection string with your actual database credentials.

Real use case: Investigating a production issue where you need to understand the data model. Ask Claude "what indexes exist on the orders table" or "show me the last 10 records where status is 'failed'." Claude runs read-only queries and explains the results.

Caveats: Read-only by design. The server cannot execute INSERT, UPDATE, DELETE, or DDL statements. This is intentional — you do not want AI-generated SQL modifying production data without explicit review. For write operations, generate the SQL and review it before running manually.

GitHub: modelcontextprotocol/servers/src/postgres

Knowledge & Search

8. Context7 MCP

What it does: Fetches up-to-date, version-specific documentation and code examples directly from library sources at query time. Instead of relying on training data that may be months old, Context7 pulls current docs when Claude needs them.

Install command:

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

Real use case: You are using a library that released a breaking change last week. Without Context7, Claude would reference the old API from its training data. With Context7, Claude fetches the current documentation and gives you the correct, up-to-date code.

Caveats: Depends on Context7's index coverage. Major libraries (React, Next.js, FastAPI, etc.) are well-covered. Niche or private libraries may not be indexed. The server adds latency for documentation lookups, but the accuracy tradeoff is worth it for fast-moving ecosystems.

Why this matters: This is arguably the highest-impact MCP server for daily coding. Stale documentation is the single biggest source of AI coding errors, and Context7 eliminates it. If you install only one server from this list, make it this one.

npm: @upstash/context7-mcp

9. Vectara MCP

What it does: Semantic search and retrieval-augmented generation (RAG) over your own documents. Vectara indexes your data and lets Claude search it with natural language queries, returning fact-checked answers with source citations.

Install command:

pip install vectara-mcp

Then configure for Claude Desktop or Claude Code using the STDIO transport:

{
  "mcpServers": {
    "vectara": {
      "command": "python",
      "args": ["-m", "vectara_mcp"],
      "env": {
        "VECTARA_API_KEY": "your-api-key",
        "VECTARA_CORPUS_KEY": "your-corpus-key"
      }
    }
  }
}

Real use case: Your company has internal documentation spread across Confluence, Google Docs, and README files. Index it all in Vectara, then ask Claude questions about your internal systems. "What is our deployment process for the billing service?" gets an answer grounded in your actual docs, not hallucinated.

Caveats: Requires a Vectara account and API key. The free tier is generous for personal use. This is a Python-based server (not npx), so you need Python 3.11+ installed. Setup is more involved than the npx-based servers, but the payoff for teams with large internal knowledge bases is significant.

GitHub: vectara/vectara-mcp

Workflow & Automation

10. Zapier MCP

What it does: Connects Claude to Zapier's ecosystem of almost 8,000 apps — Google Sheets, Slack, Jira, HubSpot, and everything else in between. Your AI assistant can trigger workflows, read data from connected apps, and automate cross-tool operations.

Install command:

Set up through the Zapier dashboard:

  1. Go to zapier.com/mcp and create a new MCP server
  2. Choose Claude as your MCP client
  3. Add the tools (apps) you want Claude to access
  4. Copy the server URL and add it to Claude Code:
claude mcp add --transport http zapier https://actions.zapier.com/mcp/YOUR_SERVER_ID/sse

Real use case: After completing a feature, tell Claude to "create a Jira ticket for QA review and post a summary in the #releases Slack channel." One instruction, two tools, zero context switching.

Caveats: Each "tool" in Zapier MCP maps to a specific action you configure in the dashboard. You need to pre-configure which apps and actions Claude can access — it cannot discover new tools on its own. The free Zapier tier works but limits the number of tasks per month.

Documentation: help.zapier.com/mcp

Communication

11. Slack MCP

What it does: Read and send messages in Slack channels, search conversation history, and manage threads directly from Claude.

Install command:

Slack's official MCP server uses OAuth authentication. The setup varies by workspace configuration, but the general approach:

claude mcp add --transport http slack https://YOUR_SLACK_MCP_URL

After adding, authenticate via the OAuth flow when prompted.

Real use case: During an incident, ask Claude to "search the #incidents channel for messages about the payment gateway outage last Tuesday." Claude finds the relevant thread, summarizes the timeline, and helps you draft a postmortem.

Caveats: Slack's official MCP server requires a registered Slack app with admin approval. In some workspaces, this means going through IT. Community alternatives exist (like jtalk22/slack-mcp-server) that use different authentication approaches, but the official server is more reliable for production use.

Documentation: docs.slack.dev/ai/slack-mcp-server

12. Notion MCP

What it does: Read and write Notion pages, query databases, search across your workspace, and manage content directly from Claude.

Install command:

claude mcp add --transport http notion https://mcp.notion.com/mcp

Then authenticate by running /mcp in Claude Code and following the OAuth flow.

Real use case: You maintain project specs in Notion. Instead of copying requirements into your Claude conversation, ask Claude to "read the API spec from the Backend Architecture page in Notion" and start implementing directly from the source of truth.

Caveats: Notion's MCP server is one of the best-implemented HTTP transport servers available. The OAuth setup is straightforward and the tool coverage is comprehensive. The main limitation is rate limits on the Notion API side, which can slow down operations on large workspaces.

Documentation: developers.notion.com/guides/mcp

Memory & Reasoning

13. Memory MCP

What it does: Persistent memory for Claude using a local knowledge graph stored as a JSON file. Claude can remember entities, relationships, and observations across sessions — project decisions, architecture choices, your preferences, team conventions.

Install command:

claude mcp add memory -- npx -y @modelcontextprotocol/server-memory

To specify a custom storage location:

claude mcp add memory -e MEMORY_FILE_PATH=~/.claude/memory.json -- npx -y @modelcontextprotocol/server-memory

Real use case: You spend 10 minutes explaining your microservices architecture to Claude. Next session, you start fresh and have to explain it again. With Memory MCP, Claude stores the architecture as entities and relationships in a knowledge graph. Next session, it already knows your service boundaries, API contracts, and deployment topology.

Caveats: The knowledge graph is local (a JSON file on disk), which means it does not sync across machines. The memory is only as good as what Claude decides to store — you may need to explicitly tell it to "remember this" for important decisions. For team-wide shared memory, you would need to build a custom MCP server that stores to a shared backend.

npm: @modelcontextprotocol/server-memory

14. Sequential Thinking

What it does: Provides a structured thinking framework that helps Claude break down complex problems step by step, with the ability to revise, branch, and refine its reasoning as understanding deepens.

Install command:

claude mcp add sequential-thinking -- npx -y @modelcontextprotocol/server-sequential-thinking

Real use case: You ask Claude to design a database migration strategy for splitting a monolithic users table into separate auth and profile services. Instead of jumping to an answer, Sequential Thinking forces a structured approach: analyze current schema, identify dependencies, plan migration phases, consider rollback strategies, then produce the final recommendation.

Caveats: This server adds thinking overhead — responses take longer because Claude is doing more structured reasoning. That is the point. Use it for architectural decisions, complex debugging, and system design. Do not use it for simple code generation where speed matters more than depth.

npm: @modelcontextprotocol/server-sequential-thinking

Web Scraping

15. Firecrawl MCP

What it does: Web scraping and crawling that converts web pages to clean Markdown. Scrape URLs, search the web, map entire domains for URL discovery, extract structured data with schemas, and run batch operations.

Install command:

claude mcp add firecrawl -e FIRECRAWL_API_KEY=your-api-key -- npx -y firecrawl-mcp

Get your API key at firecrawl.dev/app/api-keys. The free tier includes 500 credits.

Alternatively, use the hosted URL for simpler setup:

claude mcp add --transport http firecrawl https://mcp.firecrawl.dev/your-api-key/v2/mcp

Real use case: You need to analyze a competitor's documentation structure or scrape pricing pages for a comparison article. Tell Claude to "scrape the pricing page at example.com/pricing and extract the plan names, prices, and feature lists into a structured table." Firecrawl handles the rendering (including JavaScript-heavy pages) and returns clean data.

Caveats: The free tier's 500 credits go fast if you are doing batch operations. Each scrape costs credits, and complex pages cost more. For heavy scraping workloads, you will need a paid plan. Also respect robots.txt and terms of service — Firecrawl does not bypass access restrictions.

GitHub: firecrawl/firecrawl-mcp-server

How to Pick the Right MCP Servers for Your Workflow

You do not need all 15 servers. Most developers will get 80% of the value from 3-5 carefully chosen ones. Here is how to think about it:

The Essential Starter Pack

If you are just getting started with MCP, install these three:

  1. Context7 — Eliminates stale documentation errors (the single biggest improvement to AI coding accuracy)
  2. Playwright — Browser automation for testing and visual verification
  3. GitHub — Seamless pull request and issue management without leaving your editor

For Backend Developers

Add PostgreSQL or Supabase for database interaction, and Memory for persistent context about your architecture.

For Full-Stack Teams

Add Notion for reading specs, Slack for communication context, and Zapier for cross-tool automation.

For AI Agent Builders

If you are building AI-powered systems like we do at Effloow with our 14-agent company, the full stack matters. MCP servers become the sensory layer that connects your agents to the outside world. Sequential Thinking helps with complex orchestration decisions. Memory provides continuity across sessions. And if you need something that does not exist yet, build your own MCP server — it is more straightforward than you might think.

Performance Considerations

Every MCP server adds startup time and token overhead. A server that is not being used is still consuming resources during initialization. Be deliberate about what you install:

  • Scope wisely: Use --scope user only for servers you need everywhere. Use --scope local (default) for project-specific tools.
  • Remove what you do not use: claude mcp remove <name> keeps your setup lean.
  • Check status regularly: Run /mcp in Claude Code to see which servers are connected and healthy.

Build Your Own MCP Server

The 15 servers in this list cover common workflows, but the real power of MCP is that you can build servers for your specific tools and infrastructure. Need Claude to interact with your internal deployment system? Your custom analytics API? Your company's proprietary data pipeline?

We wrote a complete tutorial on how to build a custom MCP server for Claude Code, covering everything from project setup to Docker deployment. If your workflow has a tool that you wish Claude could talk to, building an MCP server for it is the answer.

What Is Next for MCP

The MCP ecosystem is moving fast. A few trends worth watching:

  • Remote MCP servers are becoming the default. The shift from stdio (local process) to HTTP transport means servers can run in the cloud, be shared across teams, and scale independently. Notion and Supabase already use this model.
  • Enterprise adoption is accelerating. CData, K2view, and other enterprise data platforms are building native MCP support, which means MCP will start appearing in corporate IT stacks, not just developer tools.
  • Agent-to-agent communication is the next frontier. While MCP handles tool access, protocols like Google's A2A (Agent-to-Agent) handle inter-agent coordination. The two protocols are complementary, not competing — expect to see them used together in production agent systems.

The best time to start building your MCP server stack was six months ago. The second best time is now. Pick three servers from this list, install them, and notice how much less context-switching you do by the end of the week.

At Effloow, we use MCP servers as core infrastructure for our AI agent workflows. This guide reflects our hands-on experience running these tools in production. For more on our approach to AI-powered development, read how we built a company with 14 AI agents or explore our Claude Code advanced workflow guide.

This article may contain affiliate links to products or services we recommend. If you purchase through these links, we may earn a small commission at no extra cost to you. This helps support Effloow and allows us to continue creating free, high-quality content. See our affiliate disclosure for full details.

← Back to Articles