devrig + MCP Steroid for JetBrains IDEs v0.102
Give your AI coding agent a real JetBrains IDE — real refactors, inspections, and a debugger.
One devrig command connects Claude Code, Codex, or Gemini to typed refactors, inspections, the debugger, and real test runs — not just text edits.
Install devrig — one command
devrig is the primary way in: one line installs the CLI and everything it needs to run — no manual setup.
curl -fsSL https://devrig.dev/install.sh | sh
irm https://devrig.dev/install.ps1 | iex
Then register your agent: devrig install claude (or replace claude with codex or gemini). Read the devrig guide →
In your JetBrains IDE, also install the MCP Steroid plugin → JetBrains Marketplace.
What a verified run looks like
We run the same task twice — same model, same repo — once with the IDE through MCP Steroid, once as a plain shell agent (grep, sed, javac). Every result below is scored on machine-checkable evidence, not on what the agent claims about itself.
Structural search across YouTrackDB
Find every no-arg Optional.get() across a multi-module reactor. The IDE's AST-exact matcher: 1m 48s / $0.75 at the same completeness the grep agent reached in 18m 13s / $3.32.
Redundant-cast inspection on Keycloak
Which files hold a genuinely redundant cast, ground-truthed by javac -Xlint:cast? The IDE inspection: 4/4 true positives in 1m 01s; the shell agent got 3/4 in 2m 40s.
Safe rename across Keycloak
An atomic PSI rename finished in 7m 39s with a clean build; the manual shell sweep died mid-way after 18 minutes with nothing usable — the win is a verified-by-construction result, not just speed.
We publish the losses too: on that same structural search, GPT 5.5's IDE leg gave up after 1m 25s without an exact match while its own grep baseline scored an exact 13/13 in 7m 32s — the Codex flow there still needs work.
Recorded on internal CI; build IDs retained for provenance (SSR 993038190, inspection 993036643, rename 992152358, GPT SSR 993038200 — all green). Reproduce from the public tests: test-experiments and the evidence-based SemanticTaskScoring.kt.
Why it matters
A coding agent limited to reading and writing files has to guess at what the IDE already knows. devrig hands it the real thing.
Real IDE semantics, not text guesses
Your agent calls the same actions a JetBrains IDE gives a human — typed rename/extract/move refactors, inspections, the debugger, and real test runs — on your actual project model.
One bridge to the IDEs you run
A single devrig process connects your agent to the IntelliJ-family IDEs already open on your machine, across projects — or it downloads and starts a managed backend. It runs in CI too — under Xvfb on Linux, or the normal GUI on macOS/Windows.
No manual MCP wiring
devrig install <agent> registers MCP Steroid with Claude Code, Codex, or Gemini, and the installer brings its own runtime — nothing to hand-configure.
devrig is the product. MCP Steroid is how it reaches your IDE.
devrig is the CLI and local control plane you install and run. To reach real IDE semantics it talks to the MCP Steroid plugin running inside a JetBrains IDE. You install devrig; devrig uses MCP Steroid.
devrig — the CLI you install
The standalone command-line control plane. It registers your agent, bridges MCP calls to a running IDE across projects, and can download, start, and stop a managed IntelliJ backend. This is the thing you install and run.
MCP Steroid — the IDE plugin
The JetBrains IDE plugin that exposes the IDE's real semantic actions. devrig connects to it so your agent reaches typed refactors, inspections, the debugger, and test runs. Also installable directly for an IDE you already use.
MCP Steroid / Devrig is an independent open-source project, not affiliated with, sponsored by, or endorsed by JetBrains. IntelliJ IDEA, PyCharm, and other IDE names are trademarks of JetBrains s.r.o.
Proof-of-Concept & Support
MCP Steroid is built by Eugene Petrenko, with 21 years of JetBrains ecosystem experience.
We run Proof-of-Concept engagements for company-specific use cases — custom skills, internal tooling integrations, AI Agent workflows tailored to your codebase. Learn more about consulting and Proof-of-Concept options.
Beyond GitHub Sponsors, MCP Steroid also accepts in-kind support: compute tokens for evaluation runs, hardware for the multi-IDE test matrix, and licenses for benchmarking and observability tools. If you or your organisation would like to back the project or the research behind it on those terms, reach out — contact details are linked below.
We are also looking for partnerships on two fronts: (a) training models — collaborating on data, evaluation, and tuning for agent-in-the-IDE workflows — and (b) embedding devrig into your company's processes — adopting it inside your engineering and CI workflows. If either fits your organisation, get in touch.