Every "best MCP servers" list you'll find is either a raw GitHub dump with no context or a copy-paste of tool names that may not even have working packages. This one is different: every tool here has been verified against the npm registry or PyPI, and every install command has been confirmed.
The anti-intuitive truth about MCP servers: the best one for your workflow isn't the most popular — it's the one that removes the specific copy-paste habit that interrupts you most often. Start there.
How to add any MCP server to Claude Code:
# HTTP servers (most tools)
claude mcp add --transport http <name> <url>
# npm-based stdio servers
claude mcp add <name> -- npx -y <package>
# Python-based servers (requires uv)
claude mcp add <name> -- uvx <package>The /mcp command inside Claude Code shows connected servers and available tools.
These five give Claude Code direct access to the tools you're already switching between during development.
GitHub MCP (official) — Read issues, PRs, commits, and file contents. Create branches and pull requests without leaving the conversation.
claude mcp add github -- npx -y @modelcontextprotocol/server-github*Best for: "implement the feature described in issue #47 and create a PR"
*
GitLab MCP (official) — Same capabilities as GitHub MCP for GitLab repositories.
claude mcp add gitlab -- npx -y @modelcontextprotocol/server-gitlabLinear MCP (community, maintained) — Read and update Linear issues directly. Claude can pull ticket context without you summarizing it.
claude mcp add linear -- npx -y linear-mcpJira MCP (community) — Query and update Jira tickets. The same package also handles Confluence pages.
claude mcp add atlassian -- npx -y mcp-atlassianSentry MCP (official, 687 GitHub stars) — Pull error details, stack traces, and issue counts from Sentry. Useful when Claude Code needs context on a bug without you copy-pasting logs.
claude mcp add sentry -- npx -y @sentry/mcp-serverGive Claude read (and in some cases write) access to your data layer — without writing query wrappers yourself.
PostgreSQL MCP (official) — Run queries, explore schemas, and describe table relationships.
claude mcp add postgres -- npx -y @modelcontextprotocol/server-postgresSupabase MCP (official) — Full Supabase access: database, auth, and storage.
claude mcp add supabase -- npx -y @supabase/mcp-server-supabaseSQLite MCP (official, Python) — Local SQLite database access. Useful for quick data analysis during development.
claude mcp add sqlite -- uvx mcp-server-sqliteNote: requires `uv` installed (`pip install uv`).
MongoDB MCP (official) — Query and explore MongoDB collections.
claude mcp add mongodb -- npx -y @mongodb-js/mongodb-mcp-serverAirtable MCP (community) — Read from and write to Airtable bases. Useful when your team uses Airtable as a lightweight database or CMS.
claude mcp add airtable -- npx -y airtable-mcp-serverThis category has the widest range in capability. The right tool depends on whether you need a quick read, an interactive session, or production-scale automation. For a deeper comparison of the underlying architectures, see Headless Browser vs AI Agents.
TinyFish MCP (official) — Five tool categories in one install. Web Search and Content Extraction are free with no credit card (pricing). Web Automation, Batch Automation (up to 8 concurrent), and Browser Sessions via CDP consume credits. Browser runs in TinyFish's cloud — no RAM overhead, session state persists for authenticated workflows on accounts you're authorized to access.
npx -y install-mcp@latest https://agent.tinyfish.ai/mcp --client claude-codeBest for: free web search and URL extraction in Claude Code, production-scale browser automation. Existing Playwright code runs unchanged via Browser Sessions.
Quick selection guide for browser tools:
| Scenario | Tool |
|---|---|
| Debugging a UI issue in Claude Code | Playwright MCP |
| Extracting content from any URL for free | TinyFish (Content Extraction) |
| Running scrapers overnight or in CI | TinyFish (Browser Sessions) |
| Quick static page fetch | Fetch MCP |
| Local headless Chrome, Puppeteer API | Puppeteer MCP |
Playwright MCP (official, Microsoft) — Connects Claude Code to a Chromium browser running on your local machine. The right choice for development debugging, one-off form submissions, and interactive site exploration where you're present.
claude mcp add playwright -- npx -y @playwright/mcp@latestDesign boundary: one session at a time, tied to your machine being on. For production pipelines, use TinyFish Browser Sessions instead. See [Playwright MCP vs TinyFish MCP](https://tinyfish.ai/blog/playwright-mcp-vs-tinyfish-browser-api?utm_source=official_blog_TT) for the full comparison.
Fetch MCP (official, Python) — Lightweight HTTP fetch wrapper. Renders pages and returns content. Good for static pages where you don't need a full browser session.
claude mcp add fetch -- uvx mcp-server-fetchPuppeteer MCP (official) — Local headless Chrome via Puppeteer. Similar to Playwright MCP, different API surface.
claude mcp add puppeteer -- npx -y @modelcontextprotocol/server-puppeteerThese five connect Claude Code to the tools where your team manages work — so Claude can read context directly instead of you copy-pasting it.
Notion MCP (official) — Read and write Notion pages and databases. Claude can pull structured content, update properties, and create new pages.
claude mcp add notion -- npx -y @notionhq/notion-mcp-serverGoogle Drive MCP (official) — Search and read Drive files. Useful when Claude needs to reference documentation or specs stored in Drive.
claude mcp add gdrive -- npx -y @modelcontextprotocol/server-gdriveGoogle Calendar MCP (community) — Read schedule and create events. Practical for Claude Code workflows that involve scheduling or time-based logic.
claude mcp add calendar -- npx -y google-calendar-mcpObsidian MCP (community) — Connect Claude Code to your local Obsidian vault. Notes and documentation become searchable context.
claude mcp add obsidian -- npx -y mcp-obsidianTodoist MCP (community) — Create and read Todoist tasks. Useful for capture workflows where Claude helps triage action items.
claude mcp add todoist -- npx -y todoist-mcp-serverFor workflows where Claude needs to read or post to team channels. Useful when relevant context lives in threads rather than docs.
Slack MCP (official) — Read channel history, search messages, and post updates. Connect Claude Code to the context that lives in Slack threads.
claude mcp add slack -- npx -y @modelcontextprotocol/server-slackGmail MCP (community) — Read, search, and draft emails. Useful for workflows where Claude needs email context alongside code.
claude mcp add gmail -- npx -y gmail-mcp-serverDiscord MCP (community) — Read and post in Discord servers. Practical for teams using Discord for technical discussions.
claude mcp add discord -- npx -y discord-mcpThese four give Claude Code observability into your infrastructure — container state, cluster health, production metrics, and local git history.
Docker MCP (community) — Container status, logs, and basic management. Useful when Claude Code is debugging a containerized application.
claude mcp add docker -- npx -y docker-mcpKubernetes MCP (community) — Cluster queries, pod status, and resource inspection.
claude mcp add kubernetes -- npx -y kubernetes-mcp-serverDatadog MCP (community) — Pull metrics, logs, and alert status from Datadog. Useful when Claude needs production observability context.
claude mcp add datadog -- npx -y datadog-mcpGit MCP (official, Python) — Local git operations: diff, log, blame, branch inspection. Useful for Claude Code workflows that need deeper git context than the built-in tools provide.
claude mcp add git -- uvx mcp-server-gitConnect Claude Code to design specs, internal documentation, and automation workflows.
Figma MCP (community, 14K GitHub stars) — Read design specs, extract dimensions, colors, and component properties. Lets Claude Code work directly from design intent.
claude mcp add figma -- npx -y figma-developer-mcpConfluence MCP (community) — Search and read Confluence pages. Same package as Jira above (mcp-atlassian) — one install covers both.
claude mcp add atlassian -- npx -y mcp-atlassiann8n MCP (community) — Trigger and monitor n8n workflows from Claude Code. Connect your automation pipelines to your AI assistant context.
claude mcp add n8n -- npx -y n8n-mcpFilesystem MCP (official) — Read and write local files with configurable allowed directories. Useful for giving Claude access to project docs, configs, or data files outside the current repo.
claude mcp add filesystem -- npx -y @modelcontextprotocol/server-filesystem /path/to/allowed/dirThe Anthropic Directory is the official source of reviewed connectors. Any server listed there can be added with claude mcp add --transport http <name> <url>.
For community servers, search npm for mcp-server-* or *-mcp packages. Check last-published date and download count before using — community packages vary in maintenance quality. All community tools in this article were verified on 2026-05-14 and had recent publish activity at time of writing.
A practical signal: if a major vendor (Notion, Supabase, Sentry, MongoDB) publishes their own MCP server, use the official one. If not, the community package with the most recent update and highest download count is usually reliable.
What are the best MCP servers for Claude Code?
The best MCP server is the one that removes your most frequent copy-paste workflow. For most developers, that means GitHub MCP (issue and PR context), PostgreSQL or Supabase MCP (database access), and Notion or Slack MCP (team knowledge). For any workflow that requires reading or extracting content from the web, TinyFish MCP's Web Search and Content Extraction tools are free with no setup beyond the install command.
How do I install MCP servers in Claude Code?
Use claude mcp add followed by the transport type and package. For npm-based servers: claude mcp add <name> -- npx -y <package>. For Python-based servers (SQLite, Fetch, Git): claude mcp add <name> -- uvx <package>. After adding, run /mcp inside Claude Code to confirm the server connected and see available tools.
Are MCP servers for Claude Code free?
The MCP protocol itself is free and open source. Most MCP server packages are free to install. Some servers connect to paid third-party services (Notion, Airtable, Datadog require their own subscriptions). TinyFish MCP's Web Search and Content Extraction tools are permanently free — only Web Automation, Batch Automation, and Browser Sessions consume credits.
What is the difference between official and community MCP servers?
Official servers are published by the vendor itself (GitHub's server is from GitHub, Notion's from Notion, Sentry's from Sentry). Community servers are third-party packages that connect to a vendor's API. Both can work well — check the npm download count and last-published date. For production use, prefer official packages where available.
How many MCP servers can I run at once in Claude Code?
Claude Code supports multiple simultaneous MCP servers. You can run GitHub, Notion, TinyFish, and a database MCP all connected at once — they appear as separate tool namespaces in Claude's context. The practical limit is how many tool options Claude can usefully consider; most developers run 3–5 servers simultaneously.
Related reading:
If TinyFish fits your use case — no selectors, no browser management, goal-driven extraction — the free tier is the fastest way to validate it against your actual workflow.
500 free steps, no credit card required.
Pass a URL and a plain English goal. Get back structured JSON.
No credit card. No setup. Run your first operation in under a minute.
