Skip to main content
Install MCP Tools: From Connector to Verified Call
Back to Blog
AI Connectors February 24, 2026 Updated: September 30, 2026 6 min readby Matthias Meyer

Install MCP Tools: From Connector to Verified Call

Set up remote MCP in Claude, verify a first call and distinguish stdio, cloud access and provider-specific requirements.

On this page▾

An MCP server connects an AI assistant to tools and data. It can help you find contacts, retrieve knowledge or assess a website's visibility. Whether the connection works depends on the client, server and authentication.

This guide starts with a ready-made remote server. We support Claude for StudioMeyer's tools. For ChatGPT's general options, see the developer mode guide; custom GPT actions are not an MCP connection method.

Choose a useful tool#

Our tools include Memory, CRM and GEO. Memory stores knowledge, CRM works with contacts, and GEO helps assess AI visibility. Start with a connection that solves a specific task.

Use the current product page for the endpoint, authentication and plan. Not every MCP offering requires a purchase or uses the same key type. An example key in a tutorial does not grant access.

Add a remote server to Claude#

Claude supports custom remote connectors in its settings. According to Anthropic, they are available on Free, Pro, Max, Team and Enterprise. Free allows one custom connector. Team and Enterprise require an Owner to add it first.

  1. Open Claude's connector settings.
  2. Add the exact endpoint from the provider's instructions.
  3. Authenticate with the supported method, often OAuth.
  4. Enable the connection for the conversation and inspect its tools.

These remote connectors connect from Anthropic's cloud, even in Claude Desktop. The server must be reachable from there. Local stdio servers in Desktop's configuration use a separate mechanism.

Remote servers in Claude Code#

For a provider supporting an HTTP endpoint, the basic pattern is:

claude mcp add --transport http example https://mcp.example.com/mcp

The .example address is a placeholder. Use the provider's confirmed address and authentication, then inspect the connection with /mcp. For StudioMeyer, follow the relevant product page. Keep secrets out of publicly shared configurations.

Check that it works#

Start with a read tool. For example, ask the assistant to retrieve a test contact you created earlier. Compare the result with the original and inspect whether the intended server was called. An answer to “Which tools do you have?” alone does not prove working authentication or data access.

If something is missing, check the endpoint, permissions, sign-in and actual error. Some changes become visible only after reconnecting or restarting. Clients do not all refresh tools in the same way.

Grant data access deliberately#

HTTPS protects data in transit. It does not establish what the server or assistant stores afterward or uses for other purposes. Storing data is part of what Memory and CRM do. Processing by the AI provider also depends on its terms and your account settings.

MCP does not have one universal API-key method: local processes, OAuth connections and provider-specific bearer tokens differ. Use the smallest necessary permissions and revocable access. Review write actions and enable only tools you need.

After a successful read, add further functions gradually. A useful connection reliably makes a real task easier.

Matthias Meyer

Matthias Meyer

Founder & AI Director

Founder & AI Director at StudioMeyer. Has been building websites and AI systems for 10+ years. Living on Mallorca since 2011, running an AI and web design studio there: web design, AI connectors, AI systems and custom-trained models, plus three self-serve MCP servers.

Update history

  • Revised MCP installation with a concrete Claude test, transport boundaries and provider-specific requirements.
Model Context Protocol

Three more posts from the same topic cluster that show how the picture fits together:

Cluster overview: MCP vs REST API vs WebMCP: When to Use Which Protocol