Guide · Updated May 2, 2026

Top MCP Servers Every Claude Code User Should Install in 2026

Quick answer

The 10 essential MCP servers for Claude Code in May 2026: filesystem, git, github, Sprintra (memory + project workspace), claude-mem (memory-only alternative), postgres, puppeteer, sequential-thinking, slack, brave-search. Most active users run 5-10 concurrently. Install via the official MCP Registry (9,400+ servers, 85% MoM growth as of Q1 2026).

The Model Context Protocol (MCP), open-sourced by Anthropic in November 2024, has become the de-facto integration layer for AI coding agents in 2026. According to the official MCP Registry maintained by Anthropic, GitHub, Microsoft, and PulseMCP, there are now 9,400+ public MCP servers as of Q1 2026, growing 18% month-over-month. Aggregate downloads across all MCP packages crossed 97 million in late 2025. 78% of enterprise AI teams report at least one MCP-backed agent in production.

But more isn't better. Each MCP server adds tools to the agent's available action space. Install too many and the agent spends extra tokens on tool selection; install too few and it hits the limits of what it can actually do. The 10 below are the cross-cutting essentials — they're what most active Claude Code users run, ranked by install frequency and impact.

filesystem

Anthropic (official)

The default. Read, write, list, search files. Foundation for any agent that touches code.

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

https://github.com/modelcontextprotocol/servers/tree/main/src/filesystem

git

Anthropic (official)

Status, log, diff, commit, branch. The agent stops asking 'what changed?' because it can see.

claude mcp add git -- uvx mcp-server-git --repository /path/to/repo

https://github.com/modelcontextprotocol/servers/tree/main/src/git

github

GitHub (official)

PRs, issues, code search across repos, workflow runs. Critical for any team-collaboration flow.

claude mcp add github -- npx -y @github/mcp-server

https://github.com/github/github-mcp-server

Sprintra

Sprintra-io (open source MIT)

Persistent memory across sessions PLUS sprints, decisions, knowledge base, dependency graph. The project brain for AI coding agents. Zero LLM cost capture.

claude mcp add sprintra -- npx -y @sprintra/mcp

https://github.com/Sprintra-io/sprintra-mcp

claude-mem

thedotmack (open source AGPL-3.0)

Memory-only plugin. Compresses tool-call observations into a vector store. 71k+ stars; the mindshare leader if memory is all you need.

npx claude-mem install

https://github.com/thedotmack/claude-mem

postgres

Anthropic (official)

Read-only database queries. Essential for any agent that ships against real data.

claude mcp add postgres -- npx -y @modelcontextprotocol/server-postgres postgresql://user:pass@host/db

https://github.com/modelcontextprotocol/servers/tree/main/src/postgres

puppeteer

Anthropic (official)

Browser automation. The agent can navigate, click, screenshot. Useful for e2e testing and scraping.

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

https://github.com/modelcontextprotocol/servers/tree/main/src/puppeteer

Sequential Thinking

Anthropic (official)

Structured chain-of-thought tool. Forces the agent to lay out reasoning steps before acting. Better quality on complex multi-step problems.

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

https://github.com/modelcontextprotocol/servers/tree/main/src/sequentialthinking

slack

Slack (official)

Read messages, post updates, manage channels. Useful for team workflows and async-coordination flows.

claude mcp add slack -- npx -y @slack/mcp-server

https://github.com/slackapi/slack-mcp-server

#10

brave-search

Brave (official)

Web search without API rate limits or surveillance. The default answer for 'I need the agent to look something up.'

claude mcp add brave -- npx -y @modelcontextprotocol/server-brave-search

https://github.com/modelcontextprotocol/servers/tree/main/src/brave-search

How to choose between Sprintra and claude-mem

They're both on this list because they solve adjacent but distinct problems. claude-mem is a memory plugin — it captures tool-call observations and makes them searchable later. Sprintra is a project workspace with memory built in — sprints, features, decisions, dependencies, knowledge base, plus zero-LLM-cost capture.

Pick claude-mem if you only want session memory inside Claude Code, you're solo, and AGPL-3.0 isn't a procurement issue for your work environment. Pick Sprintra if you also want the rest of the project state — and especially if you care about the LLM cost of capture (Sprintra is zero; claude-mem charges per observation via ChromaDB compression).

Read the full Sprintra vs claude-mem comparison for a feature-by-feature breakdown.

Bonus: domain-specific servers worth installing when you use them

stripe-mcp-server — manage Stripe billing, customers, payments from inside the agent.
linear-mcp-server (official) — read and write Linear issues; Linear ships its own MCP server as of March 2026.
atlassian-mcp-server (official) — Jira and Confluence access for enterprise teams.
cloudflare-mcp-server — manage Cloudflare Workers, KV, R2 from the agent.
notion-mcp-server — Notion pages and databases.
obsidian-mcp — local Obsidian vault access.
sentry-mcp-server — Sentry errors, releases, performance.

Frequently asked questions

What is an MCP server?

An MCP server (Model Context Protocol server) is a process that exposes tools, resources, and prompts to an AI agent like Claude Code, Cursor, or Codex. The agent connects to the server via stdio or HTTP and can call tools to read files, query databases, post to Slack, run web searches, manage project artifacts, and more. As of May 2026, the official MCP Registry lists 9,400+ servers with 85% month-over-month growth.

How many MCP servers should I install in Claude Code?

Most active users run 5-10 MCP servers concurrently. Each adds tools to the agent's available action space; too many can dilute attention and slow tool selection. A typical productive setup: filesystem + git + GitHub + a memory/PM server (Sprintra or claude-mem) + a database server (Postgres or SQLite) + a search tool (Brave or Perplexity) + Sequential Thinking. Add domain-specific servers (Slack, Linear, Stripe) only when you actively use them.

Are MCP servers safe to install?

Most official Anthropic-published servers (modelcontextprotocol/* on GitHub) are safe and audited. Third-party servers vary widely. Best practice: (a) install from the official MCP Registry or known publishers like Anthropic, GitHub, Stripe, Cloudflare; (b) review the server's README and source before granting filesystem or network access; (c) prefer servers with public security audit history; (d) be cautious of servers that request unrestricted shell access without scoped permissions.

What's the difference between an MCP server and a Claude Code plugin?

An MCP server is a standardized protocol-compliant process that any MCP client (Claude Code, Cursor, Codex, Antigravity, Gemini CLI) can connect to. A Claude Code plugin is specific to Claude Code's plugin system and includes hooks (SessionStart, UserPromptSubmit, Stop) plus optional MCP servers. Plugins are tighter Claude Code integration; MCP servers are cross-IDE portable.

Sources

Add Sprintra to your MCP setup

The project brain for AI coding agents. Memory + sprints + decisions + KB. Free OSS. 30-second install.

Install in 30 seconds

Last updated May 2, 2026. Server data verified against the official MCP Registry and each project's GitHub repository.