Category · Claude MCP servers
Claude MCP Servers: How to Connect Them and Which to Install First
A Claude MCP server extends Claude beyond conversation into your actual systems — repositories, databases, browsers, search, and project tools. Anthropic created the Model Context Protocol and Claude was its first client, so support here is deeper than anywhere else: Claude Desktop, Claude Code, and the API all speak it. This page covers how to connect servers to Claude in 2026 and which ones are worth installing before anything else.
What this category covers
MCP servers for Claude are tool providers that Claude can call during a conversation. Each server exposes a set of named tools with typed inputs; Claude decides which to call based on what you have asked for. Servers come in two shapes — local processes launched on your machine over stdio, and remote endpoints reached over the network. Claude Desktop and Claude Code both support each, configured through an MCP settings file or, for Claude Code, the claude mcp command. Our ranked roundup of the best MCP servers for Claude goes deeper on individual picks.
How it works
Add the server to your MCP configuration with its command and arguments, or its remote URL. On startup Claude discovers the tools and their schemas. From then on, tool selection is Claude's decision — which is why project-level instructions matter: telling Claude in your project rules to consult a specific server for a specific kind of task dramatically increases how often it actually does. Write operations should be gated behind approval, and Claude Code will prompt before running them by default. See also our list of the best MCP tools for Claude.
Buying guide
Install fewer servers than you think you need. Every tool a server exposes occupies context window and enters the pool Claude chooses from, and past roughly a dozen tools selection accuracy measurably degrades — the classic symptom is Claude reaching for the wrong server or ignoring a tool entirely. Start with GitHub and Context7 if you write code, add Filesystem if you use Claude Desktop rather than Claude Code, and add exactly one search server. Configure servers per-project rather than globally where your client allows it, so a Postgres connection only exists in the project that needs it. Finally, write explicit project rules about when to use which server — MCP gives Claude the capability, but instructions are what make it reach for the right one reliably. Browse the wider set of Claude MCP servers for more options.
The Tools, Ranked
Repos, issues, and pull requests. The first server most developers install — it turns Claude from a code generator into something that can read the actual codebase.
Read and write local files inside directories you allow. Essential for Claude Desktop; Claude Code has native file access, so this matters more for the desktop client.
Version-specific library docs fetched at request time. The single most effective fix for Claude writing against outdated APIs.
Microsoft's server for driving a real browser — navigation, form filling, scraping, and end-to-end testing.
Query and manage a Supabase project, including Postgres, auth, and storage, without leaving the conversation.
Real-time search built for agents, with 1,000 free searches a month. Fills the gap between Claude's training cutoff and today.
Dynamic step-by-step problem decomposition for tasks that benefit from explicit planning rather than a single pass.
Read and update Linear issues and projects, so Claude can close the loop between writing code and updating the ticket.