Best MCP Servers for Claude Code — Hand-Curated 2026 List
Hundreds of MCP servers exist. Most are demos. This is the short list we keep installed: a handful of Anthropic-maintained reference servers, plus two community servers that have earned their place.
Background: what’s official in 2026
Anthropic maintains 7 reference MCP servers as of 2026:
- Everything — test/reference server with prompts, resources, and tools
- Fetch — web content fetching and conversion for LLM use
- Filesystem — secure file operations with configurable access controls
- Git — read, search, and manipulate Git repositories
- Memory — knowledge-graph-based persistent memory
- Sequential Thinking — structured multi-step reasoning helper
- Time — timezone and time conversion
In 2025 Anthropic archived 13 other reference servers (GitHub, Slack, Postgres, Google Drive, Brave Search, Sentry, SQLite, Puppeteer, EverArt, AWS-KB, Redis, Google Maps, GitLab). Those still exist at modelcontextprotocol/servers-archived but are now vendor or community-maintained.
The short list
1. @modelcontextprotocol/server-filesystem
The single most useful server. Sandboxed file access to whitelisted directories outside the current repo. Common use: let Claude Code coordinate between a client repo and your shared toolbox in ~/Tools/.
2. @modelcontextprotocol/server-git
Structured git access. Better than raw Bash for blame, diff, and rebase operations — Claude is less likely to construct invalid command strings.
3. @modelcontextprotocol/server-time
Sounds trivial. Isn’t. In long sessions, Claude often hallucinates “as of November 2024…” comments that should say 2026. Adding the time server gives the model an authoritative current time and stops most of these.
4. @modelcontextprotocol/server-memory
The knowledge-graph memory server. Different from Claude Code’s built-in CLAUDE.md / auto-memory — this is a structured graph of entities and relations that persists across sessions. Worth trying if you do long-running research-style projects.
5. mcp-omnisearch (community)
Scott Spence’s mcp-omnisearch bundles Tavily, Brave, Kagi, Exa, Linkup, Firecrawl, and GitHub search into one MCP server. Drop in whichever API keys you have — it auto-detects what’s available and only exposes those providers. Free tiers on Tavily/Brave/Kagi are enough to get started.
What we removed
We’ve tried a lot of community MCP servers and uninstalled most. Common reasons:
- Slow startup. A 3-second-startup MCP server slows down every Claude Code session.
- Permission scope too wide. Some servers want filesystem-wide access by default.
- Bug-ridden or unmaintained. Quick to check: when was the last commit?
- Duplicates of built-in tools. Many “shell” or “browser” servers re-implement what Claude Code already has natively.
A good rule of thumb: if a server doesn’t pay rent in its first session, uninstall it.
Note on archived servers
If you previously relied on mcp-server-postgres or mcp-server-github etc., they’re still functional — just now living in modelcontextprotocol/servers-archived and maintained by the broader community / their original vendors. Use at your own discretion; treat them as you would any non-Anthropic-blessed server.
Related
Was this helpful?
Related reading
- AI coding agents keep wiping production databases — the fix isn't a smarter model, it's the permission model
- Claude Code's 33,000-token startup tax — where it goes, whether it's waste, and what to cut
- GhostApproval, explained — the symlink flaw that lets a malicious repo trick AI coding agents into writing outside your workspace
Reviews independently produced · Editorial policy
Read more reviews →