Awesome MCP Servers

85 in-depth guides to the MCP servers people actually run — what each one does, the tools it exposes, the connection details you will need, the best AI model to drive it, and how to test it free in your browser before you trust it with a real credential.

Code & developer tools

Repositories, issues, errors and containers — the servers that let an agent read and act on the work itself.

GitHub MCP Server

Hosted endpoint

Drive GitHub repos, PRs and issues with an AI agent.

Read the guide

Git MCP Server

Runs locally (stdio)

Read history, stage and commit in a local repository.

Read the guide

GitLab MCP Server

Hosted endpoint · Official

Review merge requests, read pipeline logs and search code from an MCP client.

Read the guide

Sentry MCP Server

Hosted endpoint

Triage issues, stack traces and releases from chat without opening ten browser tabs.

Read the guide

Docker MCP Server

Runs locally (stdio)

List containers, inspect images and manage your Docker environment with AI.

Read the guide

Filesystem MCP Server

Runs locally (stdio)

Give an AI agent scoped read and write access to local files.

Read the guide

Context7 MCP Server

Hosted endpoint · Official

Give your AI agent up-to-date, version-specific library docs.

Read the guide

Playwright MCP Server

Runs locally (stdio)

Give AI models real browser control — navigate, click, fill forms and screenshot any page.

Read the guide

Chrome DevTools MCP Server

Runs locally (stdio)

Let an agent open DevTools on a live page and debug it.

Read the guide

Everything MCP Server

Runs locally (stdio)

The reference server that exercises every part of the MCP spec.

Read the guide

Hugging Face MCP Server

Hosted endpoint · Official

Search models, datasets and Spaces — and run Spaces as tools.

Read the guide

DeepWiki MCP Server

Hosted endpoint · Official

Ask questions about any public GitHub repository and get answers grounded in its generated wiki.

Read the guide

Microsoft Learn MCP Server

Hosted endpoint · Official

Search and read official Microsoft and Azure documentation and code samples from a conversation.

Read the guide

Postman MCP Server

Hosted endpoint · Official

Manage workspaces, collections, environments, mocks and API specs from a conversation.

Read the guide

Databases & data warehouses

Query your own data in plain language. Read-only modes and scoped credentials matter more here than anywhere else.

Vector search & agent memory

Retrieval and recall — the servers an agent uses to remember things and to find them again by meaning.

Web search & scraping

Getting live information into a model. Everything here returns untrusted text, so keep it away from tools that can write or spend.

SEO & search visibility

Keyword, backlink and ranking data. Every call here bills against a plan quota, so constrain prompts with explicit numeric limits.

Cloud & infrastructure

Deploys, clusters and observability. Start every one of these read-only before you widen the permissions.

Project & knowledge work

Where teams keep their plans and documents — the servers that give an agent the context behind the code.

Design & content

Files, templates and decks an agent can search, generate and export without opening the editor.

Business, payments & growth

Revenue and customer systems. These are the servers where a wrong write costs real money, so read-only keys are the default.

Social media management

Scheduling, publishing and media across social channels. Posts go live on real accounts, so start with a read-only key and preview before publishing.

More servers

Everything else with a full guide on the site.

Datadog MCP Server

Hosted endpoint · Official

Query metrics, logs, traces and monitors from a conversation instead of a dashboard.

Read the guide

New Relic MCP Server

Hosted endpoint · Official

Ask NRQL questions in plain English and get entity, alert and APM data back.

Read the guide

Honeycomb MCP Server

Hosted endpoint · Official

Query high-cardinality traces and let an agent chase the outlier, not the average.

Read the guide

Amplitude MCP Server

Hosted endpoint · Official

Ask product analytics questions — funnels, retention, segments — without building the chart.

Read the guide

Box MCP Server

Hosted endpoint · Official

Search, read and organise enterprise content with an agent that respects Box permissions.

Read the guide

PlanetScale MCP Server

Hosted endpoint · Official

Query branches, inspect schema and read performance insights on a serverless MySQL fleet.

Read the guide

Sanity MCP Server

Hosted endpoint · Official

Query and edit structured content with GROQ, on a CMS built to be read by machines.

Read the guide

DataForSEO MCP Server

Hosted endpoint · Official

Raw SERP, keyword and backlink APIs priced per call, for agents that assemble their own analysis.

Read the guide

Webflow MCP Server

Hosted endpoint · Official

Read and update CMS collections, pages and site structure without opening the designer.

Read the guide

Clay MCP Server

Hosted endpoint · Official

Enrich and research accounts and contacts from a conversation, across dozens of data sources.

Read the guide

Cashfree MCP Server

Hosted endpoint · Official

Query payments, orders, settlements and refunds on India’s payment rails.

Read the guide

How to pick an MCP server

Fewer servers beat more servers. Every tool you connect enters the model's tool list, and selection accuracy falls as that list grows. Two web-search servers connected at once is worse than one, because the model now has to guess which of two near-identical tools to call. Connect what the task needs and disconnect the rest.

The credential is the boundary. An MCP server does exactly what its token permits — nothing in the protocol prevents a destructive call. A read-only database user, a fine-grained GitHub token scoped to one repo, a Kubernetes context bound to a narrow role: these are real controls, and a polite instruction in a system prompt is not.

Anything that reads the web returns untrusted text. A scraped page or a fetched document can contain text written to look like an instruction to your agent. Keep web-reading servers separate from servers that can write, deploy or spend, and read the MCP security guide before combining them.

Test with your own credential first. Most problems show up in the first ten minutes: a missing scope, a tool the model never picks because its description is vague, a response too large for the context window. Run the server in MCP Agent Studio and watch the tool calls before you automate anything.

MCP servers: common questions

What is an MCP server?

An MCP server is a program that exposes tools, resources and prompts over the Model Context Protocol so any compatible AI client can use them. The server owns the integration — the API calls, the auth, the data shaping — and the model just calls tools. That is why one Notion server works in Claude, Cursor, VS Code and a custom agent without being rewritten for each.

Which MCP servers are worth installing first?

For coding work, the filesystem and GitHub servers cover most of it. For research, pick one web server — Exa, Tavily or Firecrawl — rather than all three, since overlapping tools make models choose worse. Beyond that, install the server for the system you actually spend your day in: Postgres, Linear, Notion or Slack.

What is the difference between a hosted and a local MCP server?

A hosted server runs on the vendor's own infrastructure and you connect to a URL — 51 of the servers here work that way, usually with OAuth or a bearer token. A local server runs on your machine over stdio, typically launched with npx, uvx or docker. Local is the only option when the server needs your filesystem, your kubeconfig or a database that is not reachable from the internet.

Are MCP servers safe to connect to real accounts?

The protocol gives an agent whatever the credential allows, so the credential is the security boundary. Use a scoped, read-only token first, prefer a staging or replica target, and watch the actual tool calls before widening access. Web-facing servers add prompt injection to the picture: scraped or fetched text can contain instructions aimed at your agent.

How do I test an MCP server before wiring it into an agent?

Open it in MCP Agent Studio, connect with your own credential and run a few real prompts against 60+ models. You see every tool call and result before anything is automated, which surfaces the common problems — a missing scope, a tool the model never picks, a response too large for the context window — in minutes rather than in production.

Do all MCP servers work with every AI model?

Any model with tool calling can drive any MCP server, but reliability varies a lot. Servers with many similarly named tools, or that need multi-step chains, separate strong models from weak ones quickly. Each guide here names a recommended model and says why, and you can compare models side by side on the same connection.

Looking for something else?

These are deep guides to the most-asked-about servers. To browse the full universe of MCP servers, check the registry — or try a ready-to-run agent template.

Awesome MCP Servers — 85 Guides With Setup and Tools | MCP Playground