Delv
IDEby JetBrains★ 4.3

JetBrains AI Assistant

JetBrains' AI Assistant added MCP support in late 2025. First-class for IntelliJ/PyCharm/WebStorm users; tight refactor integration.

Homepage MCP docs
A
Safety & Trust

Delv Safety Grade: A

Score 83/100 · assessed 2026-04-19

Maintainer95
Permissions75
Supply chain85
Transparency65
Incidents100

JetBrains AI Assistant is a first-party IDE integration from a major development tools vendor with decades of track record. The MCP client implementation sits inside the IDE, inheriting JetBrains' existing security model and update infrastructure. Configuration is manual via IDE settings, which reduces supply-chain risk compared to auto-installing packages. However, the MCP feature itself is relatively new (late 2025), documentation is thin, and there's no public source code for the MCP client layer. Permissions depend entirely on which MCP servers you configure: the host itself can invoke any tool the server exposes, so filesystem writes, shell execution, or network calls are possible if you connect a server offering them. JetBrains' reputation and professional support are strong positives, but the closed-source nature and nascent MCP integration mean less community scrutiny than open alternatives.

Green flags

  • Major vendor (JetBrains) with 20+ year track record in developer tools
  • Distributed via official JetBrains Toolbox and plugin marketplace
  • Manual server configuration reduces auto-install supply-chain risk
  • Tight IDE integration allows in-editor tool invocation without context switch
  • Professional support and SLA for enterprise customers

Red flags

  • Closed-source MCP client implementation, no public audit trail
  • MCP support very recent (late 2025), limited field testing
  • Documentation sparse compared to mature JetBrains features
  • Permissions scope determined by user-configured servers, not host-enforced

Permissions requested

Read filesWrite filesOutbound networkShell executeExternal LLM call
Assessed by Delv Editorial using public metadata. Grades are advisory and update as the ecosystem changes. They do not replace your own review of permissions and code before granting an agent access to sensitive systems.

MCP capabilities

  • Tools
  • Resources
  • Prompts
  • Sampling

Platforms

macOSWindowsLinux

Config location

JetBrains settings -> Tools -> AI Assistant -> MCP Servers

Review

I spend most of my day in IntelliJ and PyCharm, so when JetBrains shipped MCP support in late 2025, I switched from Claude Desktop for most server interactions. The integration is genuinely tight. When an MCP server exposes a code-related tool, the Assistant can trigger it mid-refactor without leaving the editor. I've used this with a custom database schema server: ask a question about table relationships, get the answer inline, then immediately apply the suggested migration. No context-switching. Configuration lives in the IDE settings under Tools → AI Assistant → MCP Servers. You point it at server executables or npm packages, same as Claude Desktop. The UI is less polished than Anthropic's dedicated config screen, but it works. One quirk: if a server fails to start, the error message sometimes hides in the Event Log rather than surfacing in the settings pane. Took me twenty minutes to realise a Python path issue the first time. MCP support covers tools and resources but not prompts or sampling. That's fine for most workflows. I've connected filesystem servers, a Postgres inspector, and a custom API documentation server. The Assistant surfaces resources in its context window automatically when relevant, which feels smarter than manually attaching them in Claude Desktop. For example, ask about a function, and it pulls the relevant codebase resource without prompting. The big win is refactoring integration. If an MCP tool returns code suggestions, the Assistant can apply them directly via IntelliJ's refactor engine, preserving formatting and respecting language-specific rules. This beats copying from a chat window. I've also used it to scaffold boilerplate: ask the Assistant to generate a new REST endpoint using an OpenAPI MCP server, and it writes the handler with proper imports and error handling. Downsides: it's JetBrains-only, so if you work across editors, you'll need a second MCP client. The Assistant itself requires a subscription (part of the All Products Pack or standalone). And because it's newer than Claude Desktop, some MCP servers have edge cases that haven't been tested yet. I hit one where a server expecting stdio got confused by the IDE's process management, though that was fixed upstream. For developers already in the JetBrains ecosystem, this is the most productive MCP client I've used. It's not trying to be a general-purpose chat app. It's an IDE feature that happens to speak MCP, and that focus shows.
Verdict

If you live in IntelliJ, PyCharm, or WebStorm, this is the best MCP client for code-heavy workflows. Tight refactor integration and context-aware resource handling beat standalone chat apps. Everyone else should stick with Claude Desktop or another cross-platform option.

Good at

  • Refactor integration applies MCP tool output directly via IDE engine
  • Resources surface automatically in context when relevant
  • No editor-switching; MCP interactions happen inline
  • Supports tools and resources, which covers most practical servers
  • Config persists per project, useful for team setups

Watch out

  • JetBrains-only; no cross-editor portability
  • Requires paid subscription (All Products Pack or standalone AI Assistant)
  • Error messages sometimes buried in Event Log rather than settings UI
  • No prompt or sampling support (though rarely needed)
  • Newer than Claude Desktop, so less battle-tested with edge-case servers