AI Coding Hub

Explainer

MCP Server List: Servers Worth Adding, by Category

An MCP server gives your coding agent a new tool — access to your repo, the web, a browser, or your design files. This is a categorized list of the MCP servers worth knowing for coding agents in 2026, grouped by what they do, with a one-line note on each and where to set it up. It is client-agnostic: the same servers work in Claude Code, OpenCode, and Codex CLI.

By DK, Editor  ·  Last verified: 2026-06-20  ·  How we test  ·  Facts only — no hands-on score

01 / Definition

MCP (Model Context Protocol) is an open standard that lets a coding agent call external tools through a small server process. Instead of being limited to reading files in your project, an agent with MCP servers can search the web, open a browser, query your GitHub, or pull live library docs — whatever servers you connect.

This list groups the most useful servers by category so you can pick by need rather than scroll an alphabetical dump. Each server is added with your agent's own command (see the setup guides linked below); exact package names and URLs should always be confirmed in each server's own documentation, since they change independently of the agents.

02 / What it does

  • Docs & live context

    Context7 pulls version-pinned library documentation into the agent at query time, preventing hallucinated or outdated API usage — the most common first server for any coding agent.

  • GitHub & Git

    The GitHub MCP server gives read/write access to repos, issues, pull requests, CI status, and code search. It is now a remote server (the old npm package is deprecated); connect the hosted endpoint with a personal access token.

  • Search & web

    Exa provides semantic web search that returns full-text results rather than snippets — the most-used search server for coding agents in 2026. It needs an Exa API key.

  • Browser & testing

    Playwright drives a real browser so the agent can navigate pages, click, fill forms, and screenshot — useful for front-end work and end-to-end test tasks.

  • Terminal & files

    Desktop Commander adds persistent terminal sessions, advanced filesystem operations, and process management — handy for long-running builds or work that spans multiple projects.

  • Design

    The Figma MCP server lets the agent read your Figma designs — frames, variables, Dev Mode context — and turn them into code. See our Figma MCP setup guide for the remote and desktop options.

03 / How it works

Servers come in two shapes: local stdio servers that your agent launches as a process (most npm-packaged tools, like Context7 or Playwright), and remote HTTP/SSE servers you connect by URL (like the GitHub MCP endpoint). You register a server once with your agent and it gains those tools for the session. Keep your active set lean — 2-3 servers for a given task — because connecting more than 5-7 at once tends to overflow the tool list and degrade the agent's tool selection.

[screenshot: a coding agent session listing connected MCP servers and their tools after /mcp — e.g. context7, github, playwright. Dark terminal, 2x.]

Want the exact add commands?

This list is the what; for the how — the precise add command in Claude Code, OpenCode, and Codex CLI — see our MCP server setup guide and the curated best MCP servers page.

04 / Who it's for

  • Developers setting up a coding agent who want to know which MCP servers are worth adding first.
  • Teams standardizing a shared agent toolset and deciding which servers to enable by default.
  • Anyone hitting the limits of a bare agent (no web, no GitHub, no browser) who wants to extend it.
  • People comparing servers by category — docs, Git, search, browser, design — rather than by name.

05 / Getting started

Pick the one or two servers that match what you are building, then add them with your agent's command. The generic, client-agnostic walkthrough is in our MCP server setup guide; the curated, command-by-command picks for Claude Code, OpenCode, and Codex are in best MCP servers for coding agents; and design-file access is covered in the Figma MCP setup guide.

Our AI agents

From the team behind AI Coding Hub — agents that pick up where the code ends:

FAQ

What is the best MCP server to add first?
For most coding agents, Context7 is the highest-value first server — it injects live, version-pinned library docs so the agent stops hallucinating outdated APIs. After that, add GitHub (for repo/PR access) and a search server like Exa based on your workflow. Keep your active set to 2-3 servers at a time.
Do these MCP servers work in Claude Code, OpenCode, and Codex?
Yes — MCP is a shared standard, so the same servers work across agents; only the add command differs (claude mcp add, opencode mcp add, codex mcp add). See our MCP server setup guide for the exact command in each, verified first-hand against current versions.
How do I add an MCP server to my agent?
For a local (stdio) server you point the agent at a command, usually an npx invocation; for a remote server you provide a URL and any auth header or token. Each agent has its own add command — our MCP server setup guide walks through Claude Code, OpenCode, and Codex with copy-paste examples.
Is the GitHub MCP server still an npm package?
No. The old GitHub MCP npm package is deprecated; GitHub MCP is now a remote (hosted) server. Connect it by URL with a personal access token rather than installing a local package. Confirm the current endpoint in GitHub's MCP documentation.
How many MCP servers should I connect at once?
Keep 2-3 active for a given task. Connecting more than 5-7 at once floods the tool list faster than the model can reason about it cleanly — a phenomenon called tool bloat that leads to wrong or ignored tool calls. Register more in config but keep each session lean.

Sources: modelcontextprotocol.io/introduction · github.com/github/github-mcp-server