How to Find and Connect to Remote MCP Servers

What is Awesome Remote MCP Servers?

Awesome Remote MCP Servers is a curated directory of hosted Model Context Protocol servers. Each listed service provides an endpoint that an MCP-compatible AI client can access over a URL.

Unlike a local MCP server, a remote server runs on the provider’s infrastructure. You do not need to install an npm or Python package, launch a Docker container, or keep a background process running. You configure the endpoint, authenticate when required, and let your client discover the server’s available tools.

The directory only lists entries after their endpoints respond to an MCP initialize handshake.

Remote versus local MCP servers

The primary difference is where the server runs and how your client reaches it.

• Distribution: Local servers arrive as packages, binaries, or containers; remote servers are distributed as URLs.
• Transport: Local servers commonly use standard input and output, while remote servers use Streamable HTTP or legacy SSE.
• Runtime: A local server runs on your computer; a remote server runs on the provider’s infrastructure.
• Authentication: Local tools often use environment variables or configuration files. Remote services typically use OAuth or an API token.
• Updates: You upgrade local installations yourself, while remote providers deploy their own updates.

If you need tools that run on your own machine, use the separate awesome-mcp-servers directory.

Key features of the directory

• No local installation: Every entry points to a provider-operated endpoint.
• Broad categorization: Services cover databases, developer tools, cloud platforms, content management, finance, CRM, file storage, design, marketing, and more.
• Visible authentication requirements: Each entry identifies whether it supports anonymous access, API credentials, or OAuth.
• Practical endpoint URLs: Listings include the URL needed to configure an MCP client.
• Endpoint validation: Entries must answer the MCP initialization handshake before being listed.
• Optional quality signals: Some entries include a Glama connector badge reflecting independently scored tool-definition quality and endpoint health.

Understand the authentication legend

Check the symbol beside a server before attempting to connect:

• 🔓 No authentication: Connect anonymously without supplying credentials.
• 🔑 API key or token: Obtain a credential from the provider and supply it to your client, commonly through a Bearer authorization header.
• 🔐 OAuth: Your MCP client normally opens a browser during the first connection so you can sign in and approve access.

An OAuth connection may grant an AI client access to business data or actions. Review the provider’s authorization screen carefully and approve only the permissions your workflow needs.

Installation and setup

There is no server package to install. Setup consists of choosing an endpoint and adding it to an MCP-compatible client.

1. Open the directory and select a category relevant to your task.
2. Review each server’s description and authentication symbol.
3. Copy the listed MCP endpoint exactly, including its path such as /mcp or /mcp/sse.
4. Add the URL to your client using its command-line interface, configuration file, or connector settings.
5. Complete OAuth authorization or provide the required API token.
6. Confirm that the client connects and discovers the server’s tools.

Connect with Claude Code

The README demonstrates adding Linear as a remote HTTP server. The command assigns the connection the local name linear and points Claude Code to its endpoint:

claude mcp add --transport http linear https://mcp.linear.app/mcp

Replace linear with your preferred connection name and replace the URL with another endpoint from the directory when configuring a different provider.

Connect with an mcp.json file

Cursor, VS Code, and other clients that understand the mcp.json format can use a URL-based server definition:

{
  "mcpServers": {
    "linear": {
      "url": "https://mcp.linear.app/mcp"
    }
  }
}

The key under mcpServers is the connection name. Keep names descriptive when configuring multiple services.

Connect from Claude.ai or ChatGPT

Open the application’s settings, locate Connectors, and add the remote server URL. If the server uses OAuth, follow the browser-based sign-in and consent flow shown on first use.

Configure token authentication

For an entry marked with 🔑, obtain a token from its provider. The README notes that tokens are usually sent in an authorization header using this pattern:

Authorization: Bearer <token>

The exact place to enter this header depends on the MCP client. Do not paste real secrets into prompts, public repositories, screenshots, or shared configuration files.

Basic usage examples

After connection, the available actions depend entirely on the selected provider. Your client discovers the server’s tool definitions and can invoke them while handling a relevant request.

Search technical documentation

For public documentation access without authentication, configure an endpoint such as Astro Docs or Cloudflare Docs:

https://mcp.docs.astro.build/mcp
https://docs.mcp.cloudflare.com/mcp

You can then ask your AI client a documentation-focused question, such as “Search the Astro documentation for guidance on configuring integrations.”

Work with a database platform

The database category includes hosted integrations for MongoDB, Neon, Prisma, Supabase, and Convex. For example, the Supabase endpoint is:

https://mcp.supabase.com/mcp

Supabase uses OAuth and supports project management, SQL execution, and schema inspection. Because database tools can expose or change important data, verify the selected project and review generated operations before approving them.

Manage content

Content-management integrations include Contentful, Sanity, Storyblok, Webflow, and Wix. The Sanity endpoint is:

https://mcp.sanity.io/mcp

Once authorized, the server can query and mutate Sanity datasets and documents. A practical request could be “Find draft documents in the selected dataset.” Before requesting mutations, state the intended dataset and desired change clearly.

Use an anonymous utility

Some services require no account. For example, UI Verify audits a web page for accessibility and layout issues:

https://uiverify.ai/api/mcp

After adding the endpoint, ask the client to audit a specific page and summarize the reported issues. Anonymous access does not eliminate the need to consider whether a submitted URL or its content is sensitive.

Choosing the right server

Start with the task rather than the provider name. The directory organizes endpoints into categories including browser automation, cloud platforms, communication, CRM, databases, developer tools, finance, knowledge and memory, project management, search, security, and version control.

1. Define the action: Decide whether the AI needs to search, read, create, update, deploy, or administer resources.
2. Check authentication: Determine whether anonymous access is sufficient or whether the workflow requires an account or token.
3. Review scope: Prefer access limited to the projects, workspaces, or data needed for the task.
4. Inspect quality information: Where available, use the Glama connector badge as an additional signal for endpoint health and tool-definition quality.
5. Test safely: Begin with read-only or low-risk requests before allowing write, deployment, payment, or administrative actions.

Advanced tips

Preserve the exact transport URL

Do not remove endpoint suffixes. Most entries use a path such as /mcp, but some use legacy SSE paths such as /mcp/sse. Configure the complete URL published in the directory.

Use clear connection names

When adding several servers, choose names that identify both the provider and purpose, such as supabase-staging or cloudflare-docs. Clear names reduce the chance of invoking the wrong integration.

Separate environments

If a provider exposes sensitive or mutable resources, keep development and production connections distinct where your client and provider permit it. Explicitly identify the intended environment in prompts before asking the AI to run SQL, deploy applications, edit content, or manage infrastructure.

Apply least-privilege access

OAuth and token-based integrations may expose powerful actions. Grant only the access necessary for the workflow, avoid sharing credentials, and revoke access when the connection is no longer required.

Confirm consequential operations

Some listed servers can mutate databases, update CRM records, publish sites, send communications, or manage cloud resources. Ask the client to explain or preview a proposed operation before execution, especially when the action is destructive, public, or difficult to reverse.

Troubleshoot connection failures methodically

• Verify that the URL matches the directory entry exactly.
• Confirm that the client supports remote MCP connections and the endpoint’s transport.
• Repeat or complete the OAuth browser flow when required.
• Check that a token is valid and formatted as the provider expects.
• Consult the provider’s linked website for service-specific requirements.
• If health information is available, review the associated Glama connector page.

Conclusion

Awesome Remote MCP Servers provides a practical starting point for extending AI clients with hosted tools. Because providers operate the runtime, setup usually requires only an endpoint URL and the appropriate authentication. Choose a service by task, verify its access requirements, connect it through your client, and begin with safe, narrowly scoped requests before enabling more consequential actions.