The Best MCP Servers: 21 Picks by Category

HubSpot

HubSpot’s official MCP server lets an AI read and act on CRM records: contacts, companies, and deals. It suits revenue teams who want to query the pipeline or update records from their assistant instead of clicking through the CRM.

Salesforce

Salesforce’s MCP support is official: the Salesforce DX MCP server for development tasks, plus hosted MCP servers (now generally available) that connect an AI to your org’s data and actions. It’s the enterprise CRM pick for teams who want to query or update Salesforce in natural language.

Attio

Attio ships an MCP server that gives an AI full CRM coverage: managing deals, tasks, lists, people, companies, and notes through natural language rather than raw API calls. It’s a strong pick for modern revenue teams who run on Attio and want to operate the CRM from their assistant.

La Growth Machine

La Growth Machine is a multichannel outreach platform, and its official MCP server lets an AI build audiences, launch campaigns, and handle replies, then read pipeline generated and meetings booked from the conversation. It’s the most complete outreach server here for teams who run multichannel campaigns and want to operate them from their assistant. Hosted at https://mcp.lagrowthmachine.com/mcp; details on the La Growth Machine MCP page.

Apollo

Apollo’s official MCP server connects an AI to its B2B database for the full outbound loop: search for people and companies matching your ICP, enrich contacts with verified emails and phone numbers, create or update records, and enroll prospects into sequences. It’s the strongest pick for prospecting and data enrichment from your assistant.

Intercom

Intercom’s official MCP server ships 13 tools for its platform: reading conversations, contacts, and support data. It’s useful for support and success teams who want an assistant to triage tickets, summarize threads, or pull customer context in natural language.

Customer.io

Customer.io’s official MCP server gives an AI access to your marketing workspace: it can build segments from real customer attributes and behavior, inspect profiles, and find existing campaigns. It’s a fit for lifecycle and marketing teams who want segmentation and campaign context grounded in live data.

Peec AI

Peec AI tracks how a brand appears across AI engines, and its MCP server exposes that visibility data to an assistant: brand mentions, citations, and source coverage in LLM answers. It’s a niche but increasingly relevant server for marketing teams measuring AI search visibility (GEO).

GitHub

GitHub’s official MCP server lets an AI read repositories, search code, open issues, and submit pull requests. It’s one of the five most universally supported servers and works in every major host, covering code review, triage, and repo Q&A in one place.

Filesystem

The Filesystem server is an official reference implementation that gives an assistant scoped read and write access to local files. It’s the backbone of most coding workflows. Point it at a specific directory rather than your whole machine to keep the scope tight.

Playwright

Maintained by Microsoft, the Playwright MCP server gives an AI browser automation: navigating pages, filling forms, and scraping content. It’s the go-to for any task that needs a real browser, from end-to-end testing to pulling data off a site with no API.

Supabase

Supabase’s official MCP server connects an AI to your backend: querying the Postgres database, inspecting schema, and managing project resources. It’s a strong pick for teams building on Supabase who want to explore data or manage their project from the assistant.

Notion

Notion’s official MCP server reads and writes pages, so an assistant can pull context from your workspace or draft directly into it. It’s the most useful server for teams whose knowledge lives in Notion: notes, specs, and wikis become queryable and editable.

Slack

The community Slack MCP server (the widely used korotovsky implementation) exposes messages, channels, and search, with stdio and SSE transports. It lets an assistant read threads, summarize channels, and post updates. As a community server, review its permissions before connecting a real workspace.

Linear

Linear’s native MCP server connects an AI to its issue tracking: creating issues, updating status, and querying projects. It’s ideal for engineering and product teams who run on Linear and want to triage or file work without leaving the assistant.

Webflow

Webflow’s official MCP server lets an AI manage site content: reading and updating CMS collections, items, and pages. It’s useful for marketing and web teams who want to draft or update site content programmatically from their assistant.

Airtable

The Airtable MCP server provides read and write access to bases, with schema inspection so the AI understands your tables before touching them. It’s a strong pick for teams who use Airtable as a lightweight database or ops backend. It’s community-maintained, so check the scope you grant.

Zapier

Zapier’s MCP server is a middleware layer: instead of one integration, it connects an AI to actions across 8,000+ apps. That breadth is the appeal, one server reaches tools that don’t have a native MCP yet. The tradeoff is an extra layer between the AI and the underlying app.

n8n

n8n’s MCP server lets an AI connect to and run n8n workflows programmatically. For teams already automating with n8n, it turns those workflows into tools an assistant can trigger, bridging conversational AI and your existing automations.

Make

Make’s official MCP server (from Integromat) lets an AI trigger on-demand Make scenarios and read their results as structured data. Like Zapier, it’s a force multiplier: one server gives your assistant access to a large library of pre-built integrations and multi-app workflows.

Stripe

Stripe’s official MCP server exposes its API to an assistant for payments, customers, and billing data. It’s the reference example of a vendor wrapping its own API in MCP, useful for support and finance workflows where you query Stripe in natural language rather than the dashboard.

Method 1: Claude Code (CLI or config)

For developers, add a server from the terminal with claude mcp add. For a local server, pass its command; for a remote (hosted) server, use --transport http and the server’s URL. For example, to connect the La Growth Machine MCP server:

claude mcp add --transport http lgm https://mcp.lagrowthmachine.com/mcp

Claude Desktop and Cursor use the same idea through a JSON config entry, with the command or URL for each server.

Method 2: Customize (Claude.ai and Claude Cowork)

If you use Claude in the browser or in Claude Cowork, open the Customize panel. No files to edit, and you have two options:

• Add a server from the marketplace of connectors: pick a listed MCP (many of the servers in this article are there), authenticate, and approve it.
• Add a custom MCP server: for any server not in the directory, including hosted vendor servers or your own, paste its URL plus authentication and approve it.

Either way, the server’s tools appear in the assistant right after you connect. The La Growth Machine MCP server, for instance, connects this way in a couple of clicks.

The full La Growth Machine package (MCP + Skills)

Some vendors ship more than a server. La Growth Machine bundles its MCP server with a set of Claude Skills (packaged GTM workflows), and the whole stack installs with a single command:

curl -fsSL https://raw.githubusercontent.com/LaGrowthMachine/gtm-system/main/install.sh | sh

That pulls the open-source gtm-system repo, which contains both the MCP server and the Claude Skills. For the full walkthrough and how to build your own server, see our guide on how to build and connect an MCP server; for the difference between MCP and a plain API, see MCP vs API.