JetBrains MCP Server

The JetBrains MCP Server is a Model Context Protocol (MCP) server that has been built into the development environment since IntelliJ IDEA 2025.2. External AI clients such as Claude Code, Codex, or VS Code get access through it to the tools of the running IDE and the projects opened there. The decisive difference from plain file access: the assistant does not only work with text but with the knowledge the IDE has about the code, meaning symbols, call hierarchies, inspections, and project structure. JetBrains describes the server in its official documentation; the page names no repository of its own, and the earlier project JetBrains/mcp-jetbrains with its proxy package is considered deprecated. The details below come from the documentation as of October 22, 2026.

What can the JetBrains MCP Server do?

The documentation lists 58 tools in several groups. Analysis includes call hierarchies with analyze_calls, inspections with get_file_problems and lint_files, project builds with build_project, and queries for modules and dependencies. Search offers search_symbol, search_text, search_regex, and search_file, plus a unified skill_search. With rename_refactoring the assistant renames symbols across the project, read_file also reads files from dependencies and decompiled classes, apply_patch applies changes, and reformat_file formats. There are also Git status, run configurations, opening files in the editor, and the execute_terminal_command tool for the IDE terminal. Database tools (connections, schema, SQL queries) require the Database Tools and SQL plugin, and the debugger tools with the xdebug_ prefix require the Debugger MCP toolset plugin.

What is it particularly useful for?

The value lies where plain text search reaches its limits. If an assistant should rename a method, rename_refactoring is more reliable than search and replace because the IDE knows all usages. When troubleshooting, it can apply inspections to a file, build the project, and evaluate the reported errors instead of guessing compiler output. If you have databases connected in the project, you let the assistant look at table structures and run queries. Compared with a general file system server, this server brings IDE knowledge; compared with an AI feature built into the IDE, the choice of client stays with you.

Setup

According to the documentation the MCP server is bundled and enabled by default. Under Settings | Tools | MCP Server you check the "Enable MCP Server" box; a dialog shows what access third-party applications get to the opened projects. For detected clients such as Junie, VS Code, Claude Code, or Codex, "Auto-Configure" writes the configuration file. "Project Clients Auto-Configuration" registers the connection only in the current project. Alternatively you copy the configuration via "Copy SSE Config", "Copy Stdio Config", or "Copy HTTP Stream Config" and paste it yourself. The documentation names no fixed endpoint address; it is contained in the generated configuration. After each setup the client must be restarted.

Security: what to keep in mind

The terminal tool execute_terminal_command asks for confirmation by default. In "brave mode" (commands and run configurations without confirmation) this barrier is gone, so an assistant can run arbitrary shell commands. Enable it only in environments where that is acceptable. Under "Exposed Tools" individual tools can be switched off. For database access the documentation recommends a database user with restricted rights if the agent should only read. Retrieved code and query results enter the context of the AI client.

Costs

The server is part of the IDE, and the documentation names no separate price. The licensing model of the respective JetBrains product applies; the documentation mentions that the list_recent_sql_queries tool is not available in free subscriptions. Current terms are on JetBrains' official pages.

Who is the JetBrains MCP Server for?

It is worthwhile for developers who already work in IntelliJ IDEA or a related JetBrains IDE and want to equip an external AI client with IDE knowledge. The documentation explicitly names IntelliJ IDEA from 2025.2; the page contains no list of further supported IDEs, so check that question for your product. Start with the confirmation dialog enabled and read-only tasks before you allow the assistant terminal access and changes.

A JetBrains IDE with a built-in MCP server (IntelliJ IDEA 2025.2 or later according to the docs), an opened project, and an MCP client such as Claude Code or Codex. Database and debugger tools need additional plugins.

The JetBrains MCP Server built into the IDE gives an AI client access to search, refactoring, and further IDE tools. This guide leads from activation to the first task.

Step 1: Check the prerequisites

You need a JetBrains IDE with a built-in MCP server, IntelliJ IDEA 2025.2 or later according to the documentation, an opened project, and an MCP client such as Claude Code or Codex. Database and debugger tools additionally need the Database Tools and SQL or Debugger MCP toolset plugins.

Step 2: Enable the server

Open Settings | Tools | MCP Server and check "Enable MCP Server". The dialog shows what access third-party applications get to the opened projects. Confirm with "Enable" and "Apply".

Step 3: Connect the client

For detected clients such as Junie, VS Code, Claude Code, or Codex, "Auto-Configure" writes the configuration file. With "Project Clients Auto-Configuration" the connection applies only in the current project. Alternatively copy the configuration via "Copy SSE Config", "Copy Stdio Config", or "Copy HTTP Stream Config". Restart the client afterwards.

Limit the tools

Under "Exposed Tools" you switch off individual tools the client does not need. For databases, create a user with read-only rights if the assistant should not write.

Use brave mode deliberately

Without brave mode execute_terminal_command asks for confirmation. With brave mode, terminal commands and run configurations run without asking. Leave it off until you trust the workflow.

Step 4: First tasks

Start with reading, for example "Show the callers of this method" or "What problems does the IDE report for this file?". Changes follow afterwards, such as a project-wide rename with rename_refactoring, which you review in the version history.

No sign-in needed: the server runs in the local IDE, and the connection is made through the client configuration generated by the IDE.

Access to the projects opened in the IDE: read, search, change, and create files, refactorings, builds, run configurations, Git status, terminal, and with plugins database and debugger. Individual tools can be switched off under Exposed Tools.

The server runs locally in the IDE. Code that is read, query results, and terminal output enter the context of the connected AI client and therefore reach its model provider.

Running arbitrary shell commands in brave mode, unintended changes and refactorings in the project, write access to databases through overly broad accounts, and prompt injection via code or database content.