JetBrains AI Assistant MCP Server Setup 2026 (IntelliJ, PyCharm, WebStorm)

Key Takeaways

Where MCP configuration lives in JetBrains IDEs

Open Settings | Tools | AI Assistant | Model Context Protocol (MCP). This panel is the same across IntelliJ IDEA, PyCharm, WebStorm, and GoLand — JetBrains built AI Assistant as a shared plugin, so the MCP configuration surface doesn't differ by IDE. Click Add to create a new server entry, or paste a JSON block directly if you already have one from another client.

A second, faster path exists inside a chat session: type / and select Add Command, which opens the same configuration flow without leaving the conversation.

This is a documented feature of AI Assistant 2026.2. If your installed AI Assistant plugin is significantly older, update it before assuming a missing option is a configuration mistake rather than a version gap — JetBrains has been actively extending MCP support release over release, so an older plugin build may not expose everything described here.

The three JSON shapes JetBrains accepts

stdio (local servers) — the IDE spawns the server as a subprocess:

{
  "mcpServers": {
    "your-server-name": {
      "command": "path-or-command-to-start-server",
      "args": ["optional", "arguments"]
    }
  }
}

Streamable HTTP (remote servers) — the current recommended transport for anything not running locally:

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

SSE (legacy remote servers) — still supported, but documented as the older of the two remote transports:

{
  "mcpServers": {
    "your-server-name": {
      "url": "http://remote-server-address/sse"
    }
  }
}

JetBrains' own documentation is explicit that only these transport types are supported — a server that requires something outside stdio, Streamable HTTP, or SSE won't connect through this panel regardless of how correctly the JSON is written.

Step 1: Confirm prerequisites

Step 2: Add a local (stdio) server

Open the MCP panel, click Add, and paste a command/args block. A filesystem server is a reasonable first test since it needs no external account:

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/your/project"]
    }
  }
}

Save, and the server should appear as connected in the same panel without a full IDE restart in most cases — if it doesn't, restart the IDE and check the panel again before troubleshooting further.

Step 3: Add a remote (HTTP) server

For a hosted MCP endpoint, use the url shape instead of command/args:

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

If the remote server requires a bearer token or other header-based credential, check that specific server's own setup docs — the exact place to put auth material can differ by server, and JetBrains' JSON shape above is the transport configuration, not a universal auth wrapper.

Step 4: Verify the connection

Open an AI Assistant chat and ask a question that requires a tool from the server you just added — for the filesystem example above, something like:

List the files in this project using the filesystem MCP tools.

If AI Assistant calls the tool and returns real results, the server connected correctly. If nothing happens, move to the logs before re-editing the JSON.

When admin-managed configuration is in play

If your organization uses JetBrains IDE Services or JetBrains Central to manage AI centrally, an administrator can preconfigure the MCP servers available to your team and control whether individual developers can add their own. This matters for two situations:

Reading MCP logs

Go to Help → Show Log in Explorer (Windows/Linux) or Help → Show Log in Finder (macOS), then open the mcp folder. Each configured server gets its own log file there, which is the most reliable way to distinguish "the server never started" from "the server started but returned an error" from "the IDE never sent a request" — three failure modes that look identical from the chat panel alone.

Troubleshooting

Server added but never shows as connected. Confirm the JSON is valid — a single misplaced comma in the MCP panel's editor can silently fail to save the entry. Re-open the panel to confirm your edit actually persisted before assuming the server itself is broken.

stdio server fails immediately. Run the exact command and args from your config directly in a terminal. If it errors there, that's the actual problem — the IDE is just relaying whatever the process does, and a server that doesn't start standalone won't start invoked by the IDE either.

HTTP server connects but calls return auth errors. The transport-level JSON above (just a url) doesn't include a generic way to attach credentials. Check the specific server's documentation for how it expects auth — some accept a token in the URL, some expect an OAuth flow initiated by the client, and the two are not interchangeable.

A server you need isn't available and Add does nothing. Check whether your organization manages AI Assistant centrally through JetBrains IDE Services before spending more time on local configuration — centrally locked panels can look identical to a broken local one.

Everything looks right but nothing happens when you prompt. Check Help → Show Log in Explorer/Finder → mcp for that server's log file before re-touching the config. It's the fastest way to tell whether the process started, connected, and is actually receiving requests.

Frequently Asked Questions

Q: Does this work the same way in PyCharm and WebStorm as in IntelliJ IDEA?
A: Yes. AI Assistant is a shared JetBrains plugin, and the Model Context Protocol (MCP) panel under Settings | Tools | AI Assistant is the same across IntelliJ IDEA, PyCharm, WebStorm, and GoLand.

Q: Can I use the same MCP server config I already have for Cursor or VS Code?
A: The underlying mcpServers object with command/args for stdio or url for remote servers is close to what other clients use, but paste it into JetBrains' MCP panel rather than assuming a shared config file — JetBrains stores this through its own settings, not a file you edit directly outside the IDE.

Q: What's the difference between the Streamable HTTP and SSE transport options?
A: Streamable HTTP is the current transport for remote MCP servers; SSE (ending in /sse) is documented as the legacy remote option. Both are supported, but a server offering both should generally be configured with Streamable HTTP unless you have a specific reason to use SSE.

Q: Why can't I add my own MCP servers even though the panel is visible?
A: If your organization manages AI centrally through JetBrains IDE Services or JetBrains Central, an administrator can restrict whether individual developers can add their own servers, separately from controlling which preconfigured servers are available.

Q: Where do I look first when a configured server just isn't responding?
A: Help → Show Log in Explorer/Finder, then the mcp folder — each server gets its own log file there, which will usually show whether the process started, connected, or errored, rather than guessing from the chat panel alone.

Related Guides