Remote OpenClaw Blog
Best MCP Servers for Claude Code in 2026: Top Picks
9 min read ·
Context7 by Upstash is the best MCP server for Claude Code in 2026: it injects current, version-specific library documentation into your session, which fixes the most common cause of wrong code from any AI coding tool. This list ranks the 10 MCP servers that earn a slot in a Claude Code developer's config, each with the exact claude mcp add command and the right install scope. For a tool-agnostic ranking across every category, see the broader best MCP servers in 2026 list; this post covers what belongs in a Claude Code coding workflow.
How We Ranked These for Claude Code
This ranking optimizes for one job: making Claude Code better at shipping code. We scored each server on fit for coding workflows (docs lookup, repo operations, browser verification, database access, debugging), GitHub stars and maintenance checked against each repository on July 3, 2026, and install friction inside Claude Code. Servers that shine in other clients but add little to a coding session live in the complete ranked MCP list instead.
All install commands follow the syntax in the official Claude Code MCP documentation. To browse beyond these ten, our directory indexes 13,870 MCP servers, and Claude Code can search that directory for you.
The 10 Best MCP Servers for Claude Code
These ten servers cover documentation, repos, browsers, databases, debugging, and memory: the full loop of a real coding session.
1. Context7: current docs for every library (best overall)
Context7 by Upstash has 58,500+ GitHub stars as of July 2026 and solves the problem every Claude Code user hits: the model's training data lags the libraries you actually use. It exposes two tools, resolve-library-id and query-docs, that pull version-specific documentation into context on demand. It is #1 because it improves nearly every coding task rather than one niche, and the remote server means zero local processes.
claude mcp add --transport http context7 https://mcp.context7.com/mcpSetup note: it works without a key, but a free API key from context7.com raises rate limits. Install at user scope so every project gets fresh docs.
2. GitHub MCP Server: issues, PRs, and reviews
The official GitHub MCP Server (31,100+ stars) lets Claude Code read issues, open pull requests, and inspect CI runs without you tabbing to the browser. GitHub hosts a remote endpoint, so there is no Docker or PAT juggling for the basic flow: Claude Code handles the OAuth prompt.
claude mcp add --transport http github https://api.githubcopilot.com/mcp/Setup note: Claude Code already uses the gh CLI well; the MCP server earns its place for structured multi-repo queries and PR review workflows.
3. Playwright MCP: browser verification for frontend work
Playwright MCP from Microsoft (34,600+ stars) drives a real browser through structured accessibility snapshots instead of screenshots, so no vision model is needed. It closes the loop on frontend changes: Claude edits a component, loads the page, clicks through it, and confirms the change works.
claude mcp add playwright -- npx @playwright/mcp@latestSetup note: add it at project scope for frontend repos so the whole team gets self-verifying UI changes from the shared .mcp.json.
4. Sentry MCP: production errors in your session
Sentry MCP connects Claude Code to real production errors, stack traces, and issue context. The official Claude Code docs use it as their remote-server example, and the debugging workflow is genuinely different: "fix the top unresolved issue in Sentry" becomes a one-line prompt.
claude mcp add --transport http sentry https://mcp.sentry.dev/mcpSetup note: authentication is OAuth on first use. Install at user scope if you triage across services.
5. Postgres MCP Pro: safe database access
Postgres MCP Pro by Crystal DBA (3,000+ stars) gives Claude Code configurable read-only or read-write Postgres access plus index tuning and explain-plan analysis. Read-only mode is the killer feature: Claude inspects real schemas and data shapes while writing migrations, with no risk of mutating production.
claude mcp add --env DATABASE_URI=postgresql://user:pass@localhost:5432/db postgres -- uvx postgres-mcp --access-mode=restrictedSetup note: keep it at local scope. Connection strings are credentials and should never land in a committed .mcp.json.
6. Filesystem MCP: scoped access outside the repo
The official Filesystem MCP from the Model Context Protocol reference servers (the modelcontextprotocol/servers project has 88,000+ stars) provides allow-listed read and write access to directories beyond your working folder. Claude Code reads its own repo natively, so this matters when sessions need a docs folder or a second codebase.
claude mcp add filesystem -- npx -y @modelcontextprotocol/server-filesystem ~/shared-docsSetup note: list only the paths you want exposed; the allow-list is the security boundary.
7. Chrome DevTools MCP: performance and network debugging
Chrome DevTools MCP from Google (45,300+ stars) gives Claude Code what Playwright does not: performance traces, console errors with source-mapped stack traces, and network inspection. Use Playwright to act and DevTools to diagnose.
claude mcp add chrome-devtools --scope user -- npx chrome-devtools-mcp@latestSetup note: the repo's own docs recommend user scope, which matches how debugging tools get used across projects.
8. Serena: semantic code navigation at symbol level
Serena (26,000+ stars) wraps language servers for 40+ languages so Claude Code can find references, rename symbols, and edit at the symbol level instead of grepping text. It shines in large codebases where string search burns context fast.
claude mcp add --scope user serena -- serena start-mcp-server --context claude-code --project-from-cwdSetup note: install the binary first with uv tool install serena-agent, then run the command above; the --context claude-code flag tunes its toolset for Claude Code.
9. Memory MCP: knowledge that survives sessions
The official Memory MCP maintains a local knowledge graph of entities and relations, which gives Claude Code recall of decisions, conventions, and gotchas across sessions. It pairs well with CLAUDE.md: the file holds stable rules, the graph holds evolving project knowledge.
claude mcp add memory -- npx -y @modelcontextprotocol/server-memorySetup note: install at user scope, then add a CLAUDE.md line telling Claude when to store and retrieve memories, or the tools sit unused.
10. Sequential Thinking MCP: structured planning for hard problems
The Sequential Thinking MCP gives Claude a scratchpad tool for decomposing problems into revisable, branching thought steps. With extended thinking built into Claude Code it is no longer essential, which is why it ranks last, but it still helps on gnarly refactors where you want visible, auditable reasoning steps.
claude mcp add sequential-thinking -- npx -y @modelcontextprotocol/server-sequentialthinkingSetup note: skip it if you run Opus with extended thinking on every session; try it if you mostly run smaller, faster models.
Comparison Table
The table compares all ten picks on the three things that matter at install time: transport, credentials, and the right scope.
| Rank | Server | Best for | Transport | Auth needed | Recommended scope |
|---|---|---|---|---|---|
| 1 | Context7 | Current library docs | HTTP (remote) | Optional free key | user |
| 2 | GitHub MCP | Issues, PRs, CI | HTTP (remote) | OAuth | user |
| 3 | Playwright MCP | Browser verification | stdio (npx) | None | project |
| 4 | Sentry MCP | Production debugging | HTTP (remote) | OAuth | user |
| 5 | Postgres MCP Pro | Database inspection | stdio (uvx) | Connection string | local |
| 6 | Filesystem MCP | Files outside the repo | stdio (npx) | None | local |
| 7 | Chrome DevTools MCP | Performance and network | stdio (npx) | None | user |
| 8 | Serena | Semantic code navigation | stdio (uv) | None | user |
| 9 | Memory MCP | Cross-session memory | stdio (npx) | None | user |
| 10 | Sequential Thinking | Structured planning | stdio (npx) | None | user |
Scopes and .mcp.json: Set Up Servers the Right Way
Claude Code stores MCP servers at three scopes, and choosing the right one is the difference between a clean team setup and leaked credentials. Per the official docs, local (the default) stores the server in ~/.claude.json for the current project only, project writes a .mcp.json file at the repo root designed to be committed to version control, and user makes the server available across all your projects.
Three practical rules. Anything with credentials in the command line (like a database URI) stays at local scope, or use ${VAR} expansion in .mcp.json so teammates supply their own keys. Team-wide tooling like Playwright belongs at project scope; Claude Code prompts each teammate for approval before using servers from a checked-in .mcp.json. And run /mcp in a session to check connection status and /context to see how many tokens your servers' tool definitions consume. Our complete Claude Code MCP guide covers authentication, output limits, and troubleshooting in depth.
Limitations and Tradeoffs
MCP servers are not free wins. Every connected server adds tool definitions to your context window, so ten servers enabled at once eat a meaningful slice of your budget before you type a prompt; most developers should run three to five per project, not all ten. Stdio servers execute arbitrary code on your machine, so vet what you install and prefer official or heavily starred servers. Remote servers add a network dependency: if mcp.context7.com is down, that tool call fails mid-session. Some picks overlap; you do not need Playwright and Chrome DevTools MCP in the same project unless you both act on pages and profile them. And when a built-in feature covers the job (Claude Code already greps, reads files, and runs gh), skip the server.
Related Guides
- Best MCP Servers in 2026: The Complete Ranked List
- Claude Code MCP: The Complete Setup Guide
- How to Find MCP Servers from Inside Claude Code
- Best Claude Code Plugins in 2026
Go deeper
The operator playbooks
Production-ready PDF guides for OpenClaw and Hermes Agent — $19.99 each.
Skills for this topic
Browse all skills →Frequently Asked Questions
How do I add an MCP server to Claude Code?
Run claude mcp add in your terminal. For remote servers use claude mcp add --transport http <name> <url> ; for local servers use claude mcp add <name> -- <command> , where everything after -- is the command that starts the server. Then run /mcp inside Claude Code to verify it connected.
What is the best MCP server for Claude Code?
Context7 is the best MCP server for Claude Code in 2026. It injects current, version-specific library documentation into your session, has 58,500+ GitHub stars as of July 2026, works as a remote server with a single install command, and improves almost every coding task rather than one niche workflow.
How many MCP servers should I connect to Claude Code?
Three to five per project is the practical sweet spot. Every enabled server adds its tool definitions to your context window on every request, so connecting all ten picks at once wastes tokens. Install broadly useful servers at user scope and enable specialized ones per project.
What is the difference between local, project, and user scope in Claude Code MCP?
Local scope (the default) stores the server in ~/.claude.json for the current project only and stays private to you. Project scope writes a .mcp.json file at the repo root that you commit so the whole team shares the server. User scope makes the server available to you in every project on your machine.
Do MCP servers work with the Claude Code extension in VS Code?
Yes. MCP configuration is shared across Claude Code surfaces, so servers you add via claude mcp add or .mcp.json are available whether you run Claude Code in the terminal or through the IDE extension. Project-scoped servers still require a one-time approval per user.