Best MCP Servers for Coding in 2026

A short, verified list of the MCP servers worth adding to Claude Code for real coding work, with what each one does and the exact command to install it.

11 minIntermediate

Visit Claude Code

There are thousands of MCP servers now, and most of them you will never need. This is a tight list of the ones that actually pull their weight on a coding project, every one verified to exist with the install command shown. Add the two or three that fit your stack rather than all of them at once.

The shortlist

Each server below is added with claude mcp add. The stdio ones run locally through npx (Node) or uvx (Python); the GitHub server is a hosted remote you connect to over HTTP.

Server	What it does	Type
Filesystem	Read and write files in folders you allow	stdio (npx)
Fetch	Fetch a URL and convert it to clean markdown	stdio (uvx)
Context7	Pulls live, version-correct library docs	stdio or http
GitHub	Issues, pull requests, and code search	http (remote)
Playwright	Drives a real browser to test and scrape	stdio (npx)
Repomix	Packs a whole repo into one context blob	stdio (npx)
Postgres	Query, inspect, and tune a Postgres database	stdio (uvx)
Sequential Thinking	Structured step-by-step reasoning scratchpad	stdio (npx)
Git	Read history, diffs, and branches in a repo	stdio (uvx)

Filesystem and fetch: the basics

Filesystem lets the agent work with files outside the current project, scoped to directories you name. Fetch pulls a web page and hands back readable markdown, which is the simplest way to give the agent something it can read off the internet. Both are official reference servers.

zsh - basics

$claude mcp add filesystem -- npx -y @modelcontextprotocol/server-filesystem ~/projects

$claude mcp add fetch -- uvx mcp-server-fetch

Added stdio MCP server fetch to local config.

Context7: docs that match your version

Context7 (from Upstash) feeds the agent current documentation for a library so it stops suggesting APIs that were removed two versions ago. Run it locally through npx, or point at the hosted server. There is a guide dedicated to it and its alternatives elsewhere in this series.

zsh - context7

$claude mcp add context7 -- npx -y @upstash/context7-mcp

Added stdio MCP server context7 to local config.

GitHub: work the repo from chat

GitHub's own MCP server lets the agent triage issues, read and open pull requests, and search code across your repositories. The easiest path is the hosted remote server; add it, then run /mcp inside a session to sign in with OAuth.

terminal

claude mcp add --transport http github https://api.githubcopilot.com/mcp/
# then start Claude Code and run /mcp to authenticate

Playwright: a real browser

Microsoft's Playwright server gives the agent a controllable browser using accessibility snapshots rather than screenshots, so it can navigate pages, fill forms, and check that a UI actually works. It is the go-to for end-to-end testing and scraping tasks.

zsh - playwright

$claude mcp add playwright -- npx @playwright/mcp@latest

Added stdio MCP server playwright to local config.

Repomix, Postgres, and thinking

Repomix turns a local or remote repository into a single packaged context the agent can read in one shot, which is handy for onboarding it to an unfamiliar codebase. The Postgres MCP Pro server (crystaldba) adds safe queries, schema inspection, and index tuning. Sequential Thinking gives the model a structured scratchpad for multi-step problems.

zsh - more servers

$claude mcp add repomix -- npx -y repomix --mcp

$claude mcp add postgres --env DATABASE_URI=postgresql://user:pass@localhost:5432/app -- uvx postgres-mcp --access-mode=restricted

$claude mcp add sequential-thinking -- npx -y @modelcontextprotocol/server-sequentialthinking

Added 3 MCP servers to local config.

Less is more

Every connected server adds tools that compete for the agent's attention and your context window. Add only what a task needs, give database and filesystem servers the narrowest access that works, and remove servers you have stopped using with claude mcp remove.

Share the set with your team

Add the servers you all rely on with --scope project so they land in a committed .mcp.json. Anyone who clones the repo gets the same toolset without repeating this setup.

Result: a focused toolkit, not a junk drawer. Start with filesystem and one or two servers that match your stack, then add more only when a real task calls for them.

Related guides

Claude Code

Intermediate10 min

How to Add Any MCP Server to Claude Code

Learn the claude mcp add command, the difference between stdio and remote servers, where the config lives, and how to verify a server with /mcp.

Claude Code

Intermediate9 min

Context7 Alternatives: MCP Servers for Up-to-Date Docs

What Context7 does, when you might want something else, and the real MCP servers that keep Claude Code reading current documentation instead of stale guesses.