best MCP servers for Claude Code
Table of content
by Ray Svitla
MCP servers are how Claude Code talks to the outside world. without them, your agent is a very expensive text editor that can read files and run shell commands. with them, it becomes something that can search the web, control browsers, query databases, manage infrastructure, and occasionally order you lunch.
the ecosystem grew from “a handful of reference implementations” to “over 200 servers on the registry” in about six months. most of them work. some of them work well. here are the ones worth your time.
for what MCP actually is and why it matters, see what is MCP .
search & web
Brave Search
web search without selling your soul to Google. clean API, good results.
claude mcp add brave-search -- npx -y @anthropic-ai/mcp-server-brave-search
→ requires a Brave API key. free tier gets you 2000 queries/month.
Exa
semantic search. understands meaning, not just keywords. excellent for research.
claude mcp add exa -- npx -y exa-mcp-server
→ the difference between “find pages containing these words” and “find pages about this concept.”
Web Fetch
basic URL fetching and content extraction. markdown output.
claude mcp add web-fetch -- npx -y @anthropic-ai/mcp-server-fetch
→ essential. you want this on every install.
browser automation
Playwright MCP (Microsoft)
the official Microsoft MCP server for Playwright browser automation. page navigation, clicks, form filling, screenshots.
claude mcp add playwright -- npx -y @anthropic-ai/mcp-server-playwright
→ turns Claude Code into a browser agent. see browser agents .
Browserbase
cloud browser instances. no local Chrome needed. parallel sessions.
claude mcp add browserbase -- npx -y @browserbasehq/mcp-server
→ when you need five browsers running simultaneously and your laptop has opinions about that.
databases
PostgreSQL
direct Postgres access. queries, schema inspection, data manipulation.
claude mcp add postgres -- npx -y @anthropic-ai/mcp-server-postgres
SQLite
local SQLite database operations. great for prototyping.
claude mcp add sqlite -- npx -y @anthropic-ai/mcp-server-sqlite
Supabase
Supabase project management. database, auth, storage, edge functions.
claude mcp add supabase -- npx -y @supabase/mcp-server
version control & code
GitHub
repository management. PRs, issues, code search, actions.
claude mcp add github -- npx -y @anthropic-ai/mcp-server-github
→ the most obviously useful MCP server. create PRs, review code, manage issues — all from Claude Code.
GitLab
same energy, different platform.
claude mcp add gitlab -- npx -y @anthropic-ai/mcp-server-gitlab
Linear
issue tracking for teams that left Jira and never looked back.
claude mcp add linear -- npx -y @anthropic-ai/mcp-server-linear
file & knowledge management
Filesystem
enhanced file operations. the built-in file access is good; this extends it.
claude mcp add filesystem -- npx -y @anthropic-ai/mcp-server-filesystem
Notion
read and write Notion pages. database queries. content management.
claude mcp add notion -- npx -y @anthropic-ai/mcp-server-notion
→ pairs well with the automated PKM guide .
Google Drive
access Google Docs, Sheets, presentations. search across your drive.
claude mcp add gdrive -- npx -y @anthropic-ai/mcp-server-gdrive
Obsidian
vault access for Obsidian users. read notes, search, create links.
claude mcp add obsidian -- npx -y @anthropic-ai/mcp-server-obsidian
communication
Slack
channel messages, search, user lookups. automate your Slack life.
claude mcp add slack -- npx -y @anthropic-ai/mcp-server-slack
→ “summarize what happened in #engineering while I was sleeping” is now a real workflow.
Gmail / Google Workspace
email management. read, compose, search.
claude mcp add gmail -- npx -y @anthropic-ai/mcp-server-google-workspace
cloud & infrastructure
AWS
S3, Lambda, CloudFormation, IAM. the full AWS toolkit.
claude mcp add aws -- npx -y @anthropic-ai/mcp-server-aws
→ be careful with IAM permissions. Claude with admin access to your AWS account is a loaded footgun.
Docker
container management. build, run, inspect, logs.
claude mcp add docker -- npx -y @anthropic-ai/mcp-server-docker
Kubernetes
cluster management. pod operations, deployments, service inspection.
claude mcp add k8s -- npx -y @anthropic-ai/mcp-server-kubernetes
Terraform
infrastructure as code. plan, apply, state inspection.
claude mcp add terraform -- npx -y @anthropic-ai/mcp-server-terraform
monitoring & observability
Prometheus
metrics queries. alert management. dashboard data.
claude mcp add prometheus -- npx -y @anthropic-ai/mcp-server-prometheus
Sentry
error tracking. issue search, event details, release management.
claude mcp add sentry -- npx -y @anthropic-ai/mcp-server-sentry
PagerDuty
incident management. on-call schedules. alert routing.
claude mcp add pagerduty -- npx -y @anthropic-ai/mcp-server-pagerduty
AI & memory
Mem0
persistent memory layer. remembers context across sessions using graph + vector storage.
claude mcp add mem0 -- npx -y @mem0ai/mcp-server
→ the closest thing to giving Claude Code a hippocampus. see Mem0 deep dive .
Qdrant
vector database for semantic search over your own documents.
claude mcp add qdrant -- npx -y @anthropic-ai/mcp-server-qdrant
data & analytics
HubSpot
CRM data access. contacts, deals, companies, marketing.
claude mcp add hubspot -- npx -y @anthropic-ai/mcp-server-hubspot
Salesforce
enterprise CRM. SOQL queries, record management.
claude mcp add salesforce -- npx -y @anthropic-ai/mcp-server-salesforce
Ahrefs
SEO data. backlinks, keyword rankings, site audits.
claude mcp add ahrefs -- npx -y @anthropic-ai/mcp-server-ahrefs
the starter stack
you don’t need 30 MCP servers. you need 5 good ones. here’s the minimum for most developers:
claude mcp add web-fetch -- npx -y @anthropic-ai/mcp-server-fetch
claude mcp add brave-search -- npx -y @anthropic-ai/mcp-server-brave-search
claude mcp add github -- npx -y @anthropic-ai/mcp-server-github
claude mcp add postgres -- npx -y @anthropic-ai/mcp-server-postgres
claude mcp add playwright -- npx -y @anthropic-ai/mcp-server-playwright
add more as you actually need them. each server eats tool-description tokens from your context window. the MCP server stacking guide covers how to manage this at scale.
what this list won’t tell you
this is a catalog. catalogs are dangerous. they make you think the goal is collecting tools rather than solving problems.
the real skill isn’t knowing which MCP servers exist. it’s knowing when you need external tool access at all. most tasks Claude Code handles well are local — reading files, writing code, running tests. the moment you add a web search server, you’re inviting the entire internet’s noise into your context window. sometimes that’s necessary. often it’s a distraction.
the broader principle: any tool that connects your AI to external systems should justify its existence on every task, not just on the task you installed it for. if your Slack MCP server sits idle for 95% of sessions but eats tool-description tokens on 100% of them, it’s not free. it’s a tax.
this applies beyond Claude Code. any agent system — OpenAI’s, Google’s, whatever comes next — will face the same tradeoff between capability and noise. learn to manage that tradeoff and the specific servers become interchangeable.
the catch
every MCP server you add is a new attack surface and a new point of failure. the server runs locally (usually), but it has access to whatever you give it. the GitHub server can push to your repos. the AWS server can spin up resources. the Slack server can message your CEO at 3am.
read the sandboxing and security guide before you connect anything you care about.
Ray Svitla stay evolving