How to Add Remote MCP Tools to Claude Code Without Leaving Your Terminal

When you are already using Claude Code for day-to-day development, the next bottleneck is often not code generation itself, but all the side tasks around it: searching, preparing visuals, creating demo assets, or shortening links without breaking your flow.

This guide walks through a practical setup: adding remote MCP servers to Claude Code so the assistant can call external tools from the same terminal session where you write and review code.

What you can do

Claude Code is Anthropic's command-line programming assistant. With MCP servers attached, it can still help with coding and refactoring, but it can also reach out to tool-specific services during a task.

The source documentation lists several managed remote MCP servers, including:

• https://suno.mcp.acedata.cloud/mcp for music workflows
• https://midjourney.mcp.acedata.cloud/mcp for image generation and editing workflows
• https://flux.mcp.acedata.cloud/mcp for image workflows
• https://seedream.mcp.acedata.cloud/mcp for image workflows
• https://nanobanana.mcp.acedata.cloud/mcp for Gemini-driven image editing workflows
• https://luma.mcp.acedata.cloud/mcp for video workflows
• https://veo.mcp.acedata.cloud/mcp for video workflows
• https://seedance.mcp.acedata.cloud/mcp for video workflows
• https://serp.mcp.acedata.cloud/mcp for search workflows
• https://shorturl.mcp.acedata.cloud/mcp for shortening links

The useful part is not that these exist as separate products. The useful part is that you can wire only the servers you need into Claude Code and then ask for a complete workflow in plain language.

How it works

Claude Code can register a remote MCP server with claude mcp add. The configuration uses:

• a server name, such as serp or shorturl
• --transport http
• the remote MCP URL
• an HTTP authorization header: Authorization: Bearer <token>

A minimal command looks like this:

claude mcp add serp --transport http https://serp.mcp.acedata.cloud/mcp \
  -H "Authorization: Bearer YOUR_TOKEN"

Two details are worth calling out.

First, -H must be uppercase because lowercase -h means help. Second, if you do not pass a scope, Claude Code uses the default local scope, which only applies in the current directory. If you want the same server available across projects, add -s user. If you want a project-level setup, use a project configuration file instead.

After adding one or more servers, check what Claude Code sees:

claude mcp list

You should see the servers you registered as HTTP MCP servers.

Option 1: add only the tools you need

For most developers, I would start with a small set rather than registering every server at once. For example, if you are writing technical posts from your terminal, search and short links are enough:

claude mcp add serp --transport http https://serp.mcp.acedata.cloud/mcp \
  -H "Authorization: Bearer YOUR_TOKEN"

claude mcp add shorturl --transport http https://shorturl.mcp.acedata.cloud/mcp \
  -H "Authorization: Bearer YOUR_TOKEN"

Then, inside Claude Code, you can ask for a workflow like:

Search for recent AI video generation trends, summarize the useful references, draft a technical outline, and shorten the key links.

That is a good fit for MCP because the assistant can combine reasoning, web search, and link handling without you copying data between browser tabs.

Option 2: keep project MCP config in .mcp.json

If the workflow belongs to a repository, you can define MCP servers in a .mcp.json file at the project root:

{
  "mcpServers": {
    "serp": {
      "type": "http",
      "url": "https://serp.mcp.acedata.cloud/mcp",
      "headers": { "Authorization": "Bearer YOUR_TOKEN" }
    },
    "shorturl": {
      "type": "http",
      "url": "https://shorturl.mcp.acedata.cloud/mcp",
      "headers": { "Authorization": "Bearer YOUR_TOKEN" }
    }
  }
}

The project-level file is useful when a team wants the same MCP names and URLs. Do not commit real tokens to a public repository. In practice, keep .mcp.json out of Git or replace secrets with environment-specific values that each developer fills in locally.

The documentation also notes that project-level .mcp.json takes precedence over the global Claude configuration. That makes it a reasonable place to encode repository-specific tool choices while still allowing personal global defaults elsewhere.

A realistic builder workflow

Here is a workflow I would actually use while maintaining a developer-facing project:

1. Ask Claude Code to inspect a README or docs folder.
2. Use search via the serp MCP server to find current references for a section that depends on recent ecosystem behavior.
3. Ask Claude Code to rewrite the section with citations or summarized links.
4. Use shorturl only for links that will be pasted into release notes or social posts.
5. Run claude mcp list if a tool is not being called as expected.

For more asset-heavy work, you could add image, music, or video MCP servers later, but the operational pattern stays the same: register the HTTP endpoint, pass the Authorization header, and let Claude Code call the tool when the task requires it.

Troubleshooting checklist

If a server is not working, start with the boring checks:

• Run claude mcp list and confirm the server is present.
• Remove and re-add the server if the URL or header is wrong.
• Confirm you used uppercase -H.
• Check whether the server was added locally, globally with -s user, or through project .mcp.json.
• Make sure you did not paste a placeholder token by mistake.

The nice thing about this setup is that it is reversible. You can start with one MCP server, validate the workflow, and add more only when they remove real friction.

If you want the complete list of remote MCP URLs and the broader Claude Code examples, the original Ace Data Cloud guide is here: https://platform.acedata.cloud/documents/claude-code-mcp-all