How to Add a Remote MCP Server to JetBrains IDEs

How do I add a remote MCP server to a JetBrains IDE?

Go to Settings | Tools | AI Assistant | Model Context Protocol (MCP) and click Add. Paste a JSON config containing the server's URL into the New MCP Server dialog, click OK, then Apply. This works across the JetBrains family — IntelliJ IDEA, PyCharm, WebStorm, GoLand and the rest — with the AI Assistant plugin enabled. Typing / in the chat and choosing Add Command gets you to the same place.

Mind the build number — there are three, and they mean different things:

• 2025.1 is when the IDE could first act as an MCP client, which is what this guide is about.
• 2025.2 is when the IDE could act as an MCP server for external clients — the opposite direction of travel. This is the number most often quoted as "the MCP floor", and it is the wrong one for this task.
• 2025.3 is the one that matters here: it added Streamable HTTP and SSE alongside stdio. A remote server configured by url needs 2025.3 or later. On 2025.2, AI Assistant supports stdio only, so the url form has nothing to connect with.

Authorization is the real constraint, not the build number. JetBrains does not document an OAuth flow for AI Assistant's MCP client at all — see Authorization below. If your server requires OAuth, the build number stops being the deciding factor.

Junie, JetBrains' autonomous coding agent, is configured separately and is covered further down.

Requirements

• A JetBrains IDE with the AI Assistant plugin installed and enabled. For a remote server configured by url, 2025.3 or later; 2025.2 supports stdio only
• An active JetBrains AI tier. There is an AI Free tier with a small cloud quota, plus paid AI Pro, Ultimate and Enterprise tiers; AI Pro also comes with the All Products Pack or dotUltimate, though it cannot be activated when the IDE license comes through a License Server. AI Free is not available in the Community Editions of IntelliJ IDEA and PyCharm, in PyCharm Unified used for free, or in Android Studio — on those you need a paid tier, which is worth checking before you conclude the plugin is broken
• For an OAuth-protected server, either Node.js available for the mcp-remote proxy, or Junie

If the MCP settings page simply isn't there, one of those two is the reason — it is not a server-side problem.

Adding the server

Under Tools | AI Assistant | Model Context Protocol (MCP), click Add. The Level setting decides whether the server is available globally, across every project, or only in the current project.

There is a shortcut worth knowing: if you already use the same server in Claude Desktop, the Import from Claude button pulls that config in directly.

Choose the JSON option and paste:

{
  "mcpServers": {
    "my-server": {
      "url": "https://mcp.example.com/mcp"
    }
  }
}

Authorization

This is the part that surprises people, so it is worth being blunt about it.

AI Assistant has no browser OAuth flow for MCP servers. Across every published version of the AI Assistant MCP documentation — 2025.2, 2025.3, 2026.1, 2026.2 — there is no mention of OAuth, browser consent, or client registration, and a JetBrains tracker item covering exactly this (LLM-25012, still open as of August 2026) states that the IDE supports only a manually pasted bearer token or a local process, and that the OAuth2 flow is not supported.

So there are three honest options, depending on what your server needs.

If the server takes a static token, put it in headers:

{
  "mcpServers": {
    "my-server": {
      "url": "https://mcp.example.com/mcp",
      "headers": { "Authorization": "Bearer YOUR_TOKEN" }
    }
  }
}

If the server requires OAuth, use the mcp-remote proxy. It speaks stdio to the IDE and performs the OAuth handshake itself, in your browser, outside the IDE. This is not a legacy workaround for old builds — for an OAuth-protected server it is the working path on any AI Assistant build, including current ones:

{
  "mcpServers": {
    "my-server": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.example.com/mcp"]
    }
  }
}

Or use Junie, which does document browser authorization — see below.

To confirm the connection: the server should show a connected status in the MCP settings list, and asking AI Assistant what tools it has access to should return the server's tool list.

Junie is configured separately

Junie can work either way, and it is also the JetBrains agent that does document browser OAuth: a remote server that needs authorization appears in Junie's list with an Authorization required status, and selecting the server and choosing Authorize opens the provider's sign-in page. If you have an OAuth-protected server and you would rather not run a proxy, this is the path.

Reuse the servers you already configured. JetBrains documents a setting for exactly this: configure the server under Settings | Tools | AI Assistant | Model Context Protocol (MCP) as above, then go to Settings | Tools | AI Assistant | Agents, enable Pass custom MCP servers, and restart the IDE. Nothing else to write.

Or give Junie its own file — .junie/mcp/mcp.json in the project root, or ~/.junie/mcp/mcp.json for user level:

{
  "mcpServers": {
    "my-server": {
      "url": "https://mcp.example.com/mcp"
    }
  }
}

The canonical remote-server entry in Junie's own documentation is a url plus optional headers — there is no required type key. If you have seen "type": "streamable-http" in JetBrains documentation, that snippet points Junie at the IDE's own MCP server for debug mode, which is a different direction of travel and needs IntelliJ IDEA 2026.1.1 or newer.

What usually goes wrong

• No MCP settings page. The AI Assistant plugin isn't installed and enabled, or there is no active JetBrains AI tier. Note AI Free is not offered in Community Edition IntelliJ IDEA or PyCharm, in PyCharm Unified used for free, or in Android Studio, so on those a paid tier is required.
• The url config is rejected or ignored. The build is 2025.2, which supports stdio only, so there is nothing for a url entry to connect with. Upgrade to 2025.3+, or use the mcp-remote config above.
• The server needs OAuth and nothing happens. Expected — AI Assistant does not run an OAuth flow. Use mcp-remote, or connect the server through Junie and choose Authorize.
• The connection fails immediately. Check the URL against exactly what the server publishes, including the path and any trailing slash.
• AI Assistant works but Junie doesn't see the server. Enable Pass custom MCP servers under Settings | Tools | AI Assistant | Agents and restart the IDE, or add the server to Junie's own .junie/mcp/mcp.json.

A worked example

With a real server, using Tempreon — a hosted MCP server that carries your context across every tool you connect it to:

Tempreon authorizes over OAuth rather than a static token, so on AI Assistant it goes through the proxy:

{
  "mcpServers": {
    "tempreon": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://api.tempreon.com/functions/v1/tempreon-mcp/mcp"]
    }
  }
}

mcp-remote opens the browser approval and handles the handshake, and Tempreon supports dynamic client registration, so there are no credentials to paste. In Junie you can use the plain url form instead and choose Authorize when the server shows Authorization required. The account-holder walkthrough, including both variants, is in the JetBrains setup doc.