jbctl

English | 中文

Give your AI agent access to 41+ JetBrains IDE tools — code analysis, search, refactoring, database queries, and more.

jbctl is an Agent Skill that bridges AI agents (Claude Code, etc.) to the built-in MCP Server in WebStorm, IntelliJ IDEA, GoLand, PyCharm, and other JetBrains IDEs (2025.2+). The agent learns a discover → doctor → inspect → call workflow and can autonomously use any IDE tool.

What your agent can do

Category	Count	Examples
File Operations	12	read_file, create_new_file, replace_text_in_file, list_directory_tree
Search	5	search_text, search_regex, search_symbol, find_files_by_name_keyword
Code Analysis	2	get_symbol_info, get_file_problems
Refactoring	2	rename_refactoring, reformat_file
Project Info	3	get_project_modules, get_project_dependencies, get_run_configurations
Execution	3	execute_terminal_command, execute_run_configuration, build_project
Database	10	execute_sql_query, list_database_connections, preview_table_data
VCS & Inspection	5	get_repositories, run_inspection_kts, generate_psi_tree

Tool availability varies by IDE product and version.

Prerequisites

• A JetBrains IDE (2025.2+) with MCP Server enabled
• Enable: Settings → Tools → MCP Server → ☑ Enable MCP Server

Setup

Paste this into Claude Code to install both the CLI and the skill:

Install jbctl globally (bun i -g jbctl) and add the skill (bunx skills add https://github.com/OrdinarySF/jbctl.git).

Or install manually

1. Install the CLI

bun i -g jbctl

Or use npm: npm i -g jbctl. You can also run without installing via bunx jbctl.

2. Install the Skill

bunx skills add https://github.com/OrdinarySF/jbctl.git

This adds the skill to .claude/skills/jbctl in your project.

Alternative: Download binary

# macOS Apple Silicon
curl -fSL https://github.com/OrdinarySF/jbctl/releases/latest/download/jbctl-darwin-arm64 -o /usr/local/bin/jbctl
chmod +x /usr/local/bin/jbctl

# macOS Intel
curl -fSL https://github.com/OrdinarySF/jbctl/releases/latest/download/jbctl-darwin-x64 -o /usr/local/bin/jbctl
chmod +x /usr/local/bin/jbctl

# Linux x64
curl -fSL https://github.com/OrdinarySF/jbctl/releases/latest/download/jbctl-linux-x64 -o /usr/local/bin/jbctl
chmod +x /usr/local/bin/jbctl

Alternative: Build from source (requires Bun)

git clone https://github.com/OrdinarySF/jbctl.git
cd jbctl && bun install && bun scripts/build.ts
cp dist/jbctl-* /usr/local/bin/jbctl

Skill: Add manually

For Claude Code:

mkdir -p .claude/skills
ln -s /path/to/jbctl/skills/jbctl .claude/skills/jbctl

For other agents that support Agent Skills, point them at skills/jbctl/SKILL.md.

The skill teaches your agent the full workflow — connection check, tool discovery, schema inspection, and tool invocation — with guardrails for destructive operations and error handling.

How it works

Your Agent (Claude Code, etc.)
      |
    jbctl CLI        ← skill teaches the agent how to call this
      |
  MCP Protocol       ← auto-detects transport (Streamable HTTP / SSE)
      |
JetBrains IDE        ← code analysis, indexing, type resolution, DB access

The IDE does the heavy lifting (indexing, inspections, type resolution). jbctl gives your agent a clean CLI interface to call it. The agent doesn't need to know about MCP transports, endpoint ports, or projectPath injection — jbctl handles all of that.

Why jbctl instead of direct MCP?

JetBrains IDEs expose an MCP Server that agents can connect to directly. But direct MCP has friction in real-world setups:

Scenario	Direct MCP	jbctl
Multiple IDEs running	Manual port config per IDE	jbctl discover auto-resolves by matching your project path to the correct IDE instance
Git worktree	Broken — worktree path ≠ IDE project path, MCP config can't resolve	Works — pass -p /path/to/ide/project to route calls to the right IDE regardless of cwd
IDE restarts (port changes)	Reconfigure MCP endpoint every time	Auto-discovers the new port on each call
Agent learns new tools	Agent must understand raw MCP protocol	Skill teaches a simple CLI workflow: tools → inspect → call

In short: direct MCP works when you have one IDE, one project, no worktrees, and the port doesn't change. jbctl works everywhere else.

Key features

• Auto-discovery — jbctl discover scans for running IDEs. Matches your project path against each IDE's opened projects to find the right instance — even with multiple IDEs running simultaneously.
• Schema-first — The skill enforces inspect before call, so the agent never guesses parameter shapes.
• Safe by default — Destructive DB operations require user confirmation. Brave mode warnings when IDE dialogs will block.
• Two transports — Streamable HTTP (2026.1+) with automatic SSE fallback for older IDEs.

Example sessions

Reuse IDE database connections to query data

"Look up the last 10 orders for user 42"

The agent discovers the database connection already configured in your IDE — no DSN, no credentials, no .env:

jbctl doctor -p /your/project
jbctl call list_database_connections -p /your/project --output json
# → [{"name":"prod-readonly", ...}, {"name":"local-dev", ...}]

jbctl call execute_sql_query -p /your/project \
  --json '{"connectionName":"local-dev","query":"SELECT * FROM orders WHERE user_id = 42 ORDER BY created_at DESC LIMIT 10"}' \
  --output json

No driver install, no connection string — the IDE already has it.

Rename a function with full cross-reference safety

"Rename processOrder to handleOrder across the whole project"

The agent uses the IDE's refactoring engine, which understands types, imports, and string references — not just text replacement:

jbctl call search_symbol -p /your/project \
  --json '{"symbol":"processOrder"}' --output json
# → finds the definition and all usages

jbctl call rename_refactoring -p /your/project \
  --json '{"path":"src/services/order.ts","offset":142,"newName":"handleOrder"}' --output json
# → IDE renames the symbol everywhere: definitions, imports, type references, JSDoc

One command, zero missed references.

Known limitations

• Dynamic port — The IDE assigns a random port on startup. jbctl discover handles this automatically.
• Database tools — Only available in IDE 2026.1+. The skill includes a JDBC fallback for older versions.
• Worktree detection — Auto-discovery matches by project path. In a git worktree, pass -p pointing to the IDE's original project directory.

License

MIT