How to Use Claude Remote MCP | OAuth Authentication and API Setup Explained

For developers and end users who want to get the most out of Remote MCP on both Claude.ai and the Claude API, this guide covers the complete connection workflow with the latest 2026 specifications. From adding a connector via the Connectors UI with a single OAuth click, to calling the API directly using the mcp-client-2025-11-20 header, to understanding simultaneous connection limits by plan — everything is explained in the order you need to start implementing right away.

Article Summary by AI Chatpowered by Claude
結論powered by Claude

Remote MCP is a cloud-based MCP server exposed over HTTPS, accessible from Claude.ai, the Claude API, and mobile apps via a shared URL. Unlike Local MCP's stdio approach, it is not tied to a specific PC, meaning MCP can now be called from an iPad or smartphone, dramatically expanding its range of use.

On Claude.ai, go to Settings → Connectors, paste the URL, and complete OAuth in one click. With the Claude API, add the beta header mcp-client-2025-11-20 and specify multiple servers simultaneously in the mcp_servers array. The current version, which separates mcp_toolset into the tools array, offers flexible allowlist / denylist control at the individual tool level.

Free plans are limited to one simultaneous connection, while Pro / Max / Team / Enterprise plans support multiple connections and require an OAuth Bearer Token. Note that currently only tool calls are supported — prompts and resources are not yet available — and the old beta header mcp-client-2025-04-04 has been deprecated.

目次 (9)

What Is Remote MCP — The Fundamental Difference from Local MCP

Remote MCP is a mechanism through which Claude connects directly to an MCP server exposed as an HTTPS endpoint. While traditional Local MCP worked by "launching a stdio process on your own PC and exchanging data via standard input/output," Remote MCP has evolved into an approach where Claude "calls a cloud-based service via URL."

The official documentation states:

The server must be publicly exposed through HTTP (supports both Streamable HTTP and SSE transports). Local STDIO servers cannot be connected directly. (MCP connector — Anthropic official, as of May 2026)

This change means that iPad users without a PC, as well as mobile app users, can now use MCP. Local MCP was a closed system exclusive to Claude Desktop and Claude Code, but Remote MCP works through every entry point — Claude.ai, the official mobile app, and the Anthropic API.

The 3 Environments Where Remote MCP Works

Among Claude's product lineup, there are three entry points where Remote MCP is available.

Entry Point Connection Method Primary Use Case
Claude.ai Settings → Connectors UI End users using it in daily work
Anthropic Messages API mcp_servers array + beta header App integration / server-side processing
Claude Code claude mcp add <name> <url> Developers calling it directly from the editor

This article focuses mainly on Claude.ai and the Messages API. For MCP configuration in Claude Code, see the separate article "What Is Claude MCP — Full Feature List and 5 Connection Steps."

How to Add Remote MCP on Claude.ai

On Claude.ai, Remote MCP is managed under the name "Connectors." The steps differ depending on your plan.

Pro / Max Plans (Individual)

  1. Click your profile in the upper right → open Settings
  2. Click "Connectors" in the left menu
  3. Select "+ Add custom connector"
  4. Enter the Remote MCP server URL (must start with https://)
  5. Optionally configure an OAuth Client ID / Secret
  6. Click "Add" to proceed to the OAuth authentication screen; once you log in, it's ready to use

Team / Enterprise Plans (Owner Setup)

The owner registers a web-type custom connector at the organization level via Organization settings → Connectors → Add, and each member enables the connection under Customize → Connectors (Custom Connectors — Anthropic Help Center).

Free users can also use Remote MCP, but are limited to one simultaneous connector. A Pro plan or higher is required for cross-service access with multiple connectors.

Calling Remote MCP via the Claude API MCP Connector

Developers can call Remote MCP servers directly from the Anthropic Messages API. There is no need to implement an MCP client on the Claude side — a single request is all it takes.

You need to specify a beta header to use this feature.

curl https://api.anthropic.com/v1/messages \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "anthropic-beta: mcp-client-2025-11-20" \
  -d '{
    "model": "claude-opus-4-7",
    "max_tokens": 1000,
    "messages": [{"role": "user", "content": "What tools do you have available?"}],
    "mcp_servers": [
      {
        "type": "url",
        "url": "https://example-server.modelcontextprotocol.io/sse",
        "name": "example-mcp",
        "authorization_token": "YOUR_TOKEN"
      }
    ],
    "tools": [
      {
        "type": "mcp_toolset",
        "mcp_server_name": "example-mcp"
      }
    ]
  }'

Note that the older beta header mcp-client-2025-04-04 has been deprecated; the current version is mcp-client-2025-11-20. The old format required writing tool_configuration inside the MCP server definition, whereas the current version separates it as mcp_toolset in the tools array, offering more flexible allowlist / denylist / per-tool configuration (MCP connector official docs).

OAuth Authentication — Obtaining and Refreshing the Bearer Token

Most Remote MCP servers require OAuth authentication. When using Claude.ai, the OAuth flow is handled automatically by the UI, but when using the API, it is the caller's responsibility to obtain and refresh the token.

The official documentation specifies how to handle OAuth when using the API:

API consumers are expected to handle the OAuth flow and obtain the access token prior to making the API call, and to refresh the token as needed.

For obtaining access tokens during development, the officially provided MCP Inspector is convenient.

npx @modelcontextprotocol/inspector

After launching, select SSE or Streamable HTTP, enter the server URL, then go to "Open Auth Settings" → "Quick OAuth Flow" to proceed to the authorization screen, where your access_token will be displayed. For production environments, you will need to set up a separate mechanism to securely store and automatically refresh the token.

List of Official Remote MCP Servers Available

Major SaaS providers partnered with Anthropic have already published Remote MCP servers. The following are representative examples you can connect to with a single OAuth click from the Connectors screen in Claude.ai:

  • Notion — Page search, new page creation, database operations
  • Linear — Issue creation, status updates, team-specific search
  • Atlassian (Jira / Confluence) — Ticket creation, document search
  • Asana — Task management, cross-project search
  • Cloudflare — Workers / R2 / DNS operations
  • Plaid — Read access to financial data
  • Sentry — Error monitoring, stack trace review
  • Intercom — Customer support conversation history search

Beyond these, many other SaaS providers including Stripe, Square, PayPal, HubSpot, and Salesforce have already published Remote MCP servers. For the latest full list, refer to the Anthropic official connector directory.

Local MCP vs. Remote MCP — Which Should You Choose?

Since many use cases involve both approaches, here is a quick reference for choosing between them.

Perspective Local MCP Remote MCP
Transport stdio Streamable HTTP / SSE
Runtime environment Claude Desktop / Claude Code only Claude.ai / iPad / Web / API — all
Server implementation Local launch via npm package, etc. Always running in the cloud
Authentication OS permission-based OAuth Bearer Token
Primary use cases Local file operations, git, DB SaaS integration, team sharing
Mobile support No Yes

Local MCP is better suited for use cases involving "accessing resources on a PC," such as local file operations or git. On the other hand, if you plan to integrate with SaaS or use MCP across multiple devices, Remote MCP is the clear choice.

Limitations and Pitfalls to Watch Out For

Here is a summary of the current constraints explicitly noted in the official documentation.

  • Tool calls only: Prompts / resources / sampling as defined in the MCP specification are not supported. If you need something other than tool calls, you will need to go through the TypeScript SDK's client-side helper.
  • HTTP exposure required: Local STDIO servers cannot be connected to directly from the API. To make a stdio server accessible remotely, you need to write your own HTTP wrapper or route through a relay such as Cloudflare Workers.
  • Limited supported platforms: The API's MCP connector is available on the Claude API, Claude Platform on AWS, and Microsoft Foundry. Amazon Bedrock and Google Vertex AI are currently not supported.
  • Not eligible for ZDR (Zero Data Retention): Tool definitions and execution results exchanged via MCP are subject to Anthropic's standard data retention policy. Exercise caution when handling sensitive data.

Summary — Remote MCP Is the Standard Gateway to "Using Claude from Anywhere"

Remote MCP is the mechanism that elevates Claude from "a tool you sit down at your PC to use" to "a gateway for operating SaaS services across all your devices." For Claude.ai UI users, a single click in Connectors gets you started; for developers, a single beta header line in the Messages API does the trick. Try connecting one of your everyday services — Notion or Linear, for example — and you will immediately feel the difference from Local MCP.

参考になったら ♡
Clauder Navi 編集部
@clauder_navi

Anthropic の Claude / Claude Code を中心に、日本のエンジニア向けに最新動向と実務 を毎日発信。 運営方針 は メディアについて をご覧ください。