Best MCP Servers for Claude Code, Cursor and Codex (2026)
Plug tools into your coding agent: repos, current docs, a real browser, your database, and your error tracker.
MCP servers give a coding agent tools it does not have on its own: your GitHub repo, current library docs, a real browser, your database, and your error tracker. We rank them on how much they help a builder's daily loop, how easy they are to add to Claude Code, Cursor, and Codex, what they cost, and how tightly you can scope them. Every server here comes from the team behind the product. We did not run a benchmark. The takes stick to each vendor's docs, README, and pricing page. How we rank.
These servers plug into the coding agents and AI IDEs you already use. Supabase MCP talks to the S-tier Supabase backend, and Figma MCP feeds design to code.
Quick verdict: Best overall: GitHub MCP Server. Best for current docs: Context7. Best for browser testing: Playwright MCP.
← New releases · All tiers · Compare · FAQ
Last updated:
Some links elsewhere on ToolTier are affiliate links that may earn us a commission at no extra cost to you. None of the links on this page are, and affiliate status never affects a tool's tier.
MCP servers, compared
| Tool | Tier | Starting price | Best for |
|---|---|---|---|
| GitHub MCP Server | S | Open source (MIT) | Repos, issues, PRs, and Actions from your agent |
| Context7 | S | Free (1,000 API calls/mo) | Current library docs in context |
| Playwright MCP | S | Free and open source (Apache-2.0) | Cross-browser automation and testing |
| Chrome DevTools MCP | A | Free and open source (Apache-2.0) | Debugging and profiling in Chrome |
| Supabase MCP | A | Open source (Apache-2.0) | Agent access to your Supabase project |
| Sentry MCP | A | Developer free (1 user, MCP access) | Fixing production errors from your agent |
| Figma MCP | B | Starter free (20 tool calls/mo) | Design context for code, on a paid seat |
Starting price is the first rate on each price line. Full prices are on the cards below. Order follows the tiers. Affiliate links never change it.
S tier
GitHub MCP Server site ↗
GitHub's official server for repos, issues, pull requests, Actions runs, and code security alerts. Use the hosted remote server at api.githubcopilot.com/mcp/ with OAuth or a personal access token, or run it locally. Turn on only the toolsets you need to keep the agent's context small, add /readonly to the URL for read-only tools, and use lockdown mode to hold back public-repo content from people without push access (a best-effort guard against prompt injection, not a permission boundary). It documents setup for Claude Code, Codex, Cursor, OpenCode, Zed, and Copilot. MIT licensed. v2.1.0 shipped Nov 3. v2.2.0 (Nov 16) replaces server instructions with filtered MCP skills, adds public agent automation tools, and lets secret scanning alert tools optionally hide secrets.
Context7 site ↗
Pulls current, version-specific library docs and code examples into your agent's context, so it stops guessing at APIs from old training data. Runs as an MCP server at mcp.context7.com/mcp or as the ctx7 CLI plus a skill, and npx ctx7 setup configures Cursor, Claude Code, or OpenCode. Made by Upstash. Free is 1,000 API calls a month; after that Free users are blocked, with 20 bonus calls a day until the month resets. Pro users are never blocked and pay $5 per extra 1,000 calls.
Playwright MCP site ↗
Microsoft's browser automation server. The agent drives Chrome, Firefox, WebKit, or Edge through structured accessibility snapshots instead of screenshots, so no vision model is needed. Adding it is one line, for example claude mcp add playwright npx @playwright/mcp@latest or codex mcp add. Its own README says coding agents may do better with the more token-efficient Playwright CLI plus skills, and keeps MCP for long, stateful browser loops. v0.0.84 (Oct 31) adds isolated sessions in one browser and URL parameter redaction.
A tier
Chrome DevTools MCP site ↗
The Chrome DevTools team's server for debugging a live Chrome: performance traces with insights, network requests, screenshots, and console messages with source-mapped stack traces. It also ships a chrome-devtools CLI and skills for accessibility, LCP, and memory-leak debugging, plus a --slim mode with fewer tools. Chrome and Chrome for Testing only. Google collects usage statistics by default (opt out with --no-usage-statistics), and performance tools can send trace URLs to the CrUX API (turn that off with --no-performance-crux). v1.12.0 shipped Oct 22.
Supabase MCP site ↗
Lets your agent list tables, run SQL, apply migrations, query logs, read security and performance advisors, generate TypeScript types, and deploy Edge Functions on your Supabase project. Hosted at mcp.supabase.com/mcp with OAuth, so no token is needed. Supabase warns about prompt injection from data in your tables: scope it to one project with project_ref, add read_only=true, limit feature groups, keep tool approval on, and never hand it to your end users. Branching tools need a paid plan.
Sentry MCP site ↗
Sentry's remote server for coding agents: search errors, traces, and logs, triage issues, and read Sentry docs from Claude Code, Cursor, or VS Code. Connect with OAuth at mcp.sentry.dev/mcp and scope the URL to one org or project, which Sentry recommends. A Claude Code plugin adds a Sentry subagent. If you run the stdio version yourself, its natural-language search tools need your own LLM provider key. Sentry's free Developer plan (one user) includes MCP access.
B tier
Figma MCP site ↗
Gives your agent design context from Figma (variables, components, layout) to generate code from a selected frame, and can write native content back to the canvas (beta, free for now, usage-based later). Use the remote server at mcp.figma.com/mcp. The catch is limits: Starter plans and View or Collab seats get up to 20 read tool calls a month, while Full or Dev seats on Professional get up to 200 a day. Only clients in Figma's MCP catalog can connect, including Claude Code, Codex, Cursor, and VS Code. Nov 17: admins can keep an allowlist of approved OAuth apps, including MCP clients like Cursor and Claude Code, and review app requests.
Comparisons
FAQ
What are the best MCP servers for Claude Code?
Start with the three S-tier servers: GitHub MCP Server for repos, issues and pull requests, Context7 for current library docs, and Playwright MCP for browser testing. Add Supabase MCP or Sentry MCP (A tier) when your app uses them. Most install with one claude mcp add command.
What are the best MCP servers for Cursor?
The same three. GitHub MCP Server, Context7 and Playwright MCP all document a Cursor setup, and Context7 is in the Cursor marketplace. Every other server on this list documents Cursor too. Cursor itself is S tier on the AI IDE tier list.
Do these MCP servers work with Codex and OpenCode?
Mostly, yes. GitHub MCP Server, Playwright MCP, Supabase MCP and Figma MCP document Codex and OpenCode setup, Context7's setup command targets OpenCode, and Sentry MCP works with any client that supports remote HTTP servers. See the coding agents tier list.
Are MCP servers free?
The servers themselves mostly are. GitHub MCP Server, Playwright MCP, Chrome DevTools MCP and Supabase MCP are open source. You pay for the service behind some of them: Context7 Free is 1,000 API calls a month, Sentry's free Developer plan includes MCP access for one user, and Figma Starter is capped at 20 tool calls a month.
Playwright MCP vs Chrome DevTools MCP: which should I use?
Playwright MCP is S tier for automating Chrome, Firefox or WebKit through accessibility snapshots. Chrome DevTools MCP is A tier for debugging and profiling in Chrome. See Playwright MCP vs Chrome DevTools MCP.
Should I use an MCP server or a CLI?
For a coding agent, often the CLI. Playwright's README says its CLI plus skills is more token-efficient because it avoids loading large tool schemas and accessibility trees, and keeps MCP for long, stateful browser loops. Context7 and Chrome DevTools MCP also ship a CLI. MCP is simpler for remote services that need sign-in, like GitHub, Supabase, Sentry and Figma.
Are MCP servers safe to use?
They act with your permissions, so scope them. Supabase recommends project scoping, read-only mode and fewer feature groups, and warns about prompt injection from data in your tables. GitHub MCP Server has read-only URLs and a lockdown mode for public repos. Keep tool-call approval on for anything that can write.
How many MCP servers should I install?
Only the ones you use. Each server adds tool definitions to the agent's context. GitHub's server lets you turn on only the toolsets you need for that reason, and Chrome DevTools MCP has a --slim mode with fewer tools.
More tier lists
App Builders tier listDeploy & Hosting tier listBackend, DB & Auth tier listAI Coding IDEs tier listCoding Agents tier listAI Workflow Automation tier listAI Code Review tier listDesign → Code tier listAI APIs & Models tier list
The ToolTier Drop
New tools worth your time and tier changes, every week. No hype.