Supported coding tools
The seven agentic coding environments LEX AI supports, where each keeps its MCP config, how the agent payload is rendered per tool, and what setup writes or leaves alone.
Supported coding tools
LEX AI ships as one MCP server, lex-mcp-local, and drives whichever agentic coding tool you already use. Seven are supported. Each one gets the server registered in its own config format and the same agents, slash commands, and workspace rules rendered into its native layout, so the ten workflow modes behave the same wherever you sit.
The seven environments
Tool | Selector key | MCP config location |
|---|---|---|
GitHub Copilot in PyCharm / JetBrains IDEs |
|
|
GitHub Copilot in VS Code |
| workspace |
GitHub Copilot CLI |
|
|
Cursor |
|
|
Claude Code |
| project |
OpenAI Codex |
|
|
Windsurf |
|
|
The JetBrains entry covers the whole IntelliJ family; the Claude Code entry covers both the CLI and the Claude Desktop Code tab. pycharm-copilot, copilot-cli and windsurf are configured only in your home directory — they have no project-scoped roster.
One payload, rendered per tool
The agents, prompts, and rules are authored once and translated for each tool. The three Copilot surfaces share a dialect and receive the canonical .github tree verbatim; Cursor, Claude Code, Codex, and Windsurf each have their own. A parametrised test asserts that all seven render all ten modes and always produce that tool's workspace-rules file, so no combination is quietly unsupported.
Two differences are worth knowing:
- Named subagents — the Copilot surfaces and Claude Code can spawn one. Cursor, Codex, and Windsurf cannot, so the step agents ship to a neutral
.lex/agents/tree and the workspace rules carry a delegation protocol instead: read the agent file in full, run it as one focused sub-task, drop the working detail, report the summary. The no-skip, halt-on-ok: false, and hands-off-git rules are restated verbatim; only the spawning mechanism differs. - Scoped instruction files — six of the seven have a home for them. Codex reads only
AGENTS.md, so each scoped rule is inlined into its managed block instead.
Selecting your tools
Pass -e/--environment to lex setup-with-ai, once per tool, or change the selection later in the dashboard. lex setup-with-ai --list-environments prints the keys. With no recorded selection, pycharm-copilot is the default. Every config LEX AI writes uses the single server entry name lex-mcp-local, and legacy namespaces are rewritten onto it, because a stale namespace grants your assistant no MCP access at all.
What is written, and what is left alone
- Config writes are merge-preserving and atomic — unrelated MCP servers, unknown keys, and TOML comments and formatting all survive, and a corrupt config never blocks onboarding.
CLAUDE.mdandAGENTS.mdare never overwritten — only the region between the managed-block markers is rewritten, so your own prose stays.claude_desktop_config.jsonis read but never written..lex/ai-assets.jsonrecords what was generated — and lists only files LEX AI may delete, which is why deselecting a tool removes exactly its own assets.- A managed
.gitignoreblock covers the project-scoped configs — they holdGITHUB_TOKENandREMOTE_MCP_API_KEY, and the forward workflow runsgit add -Aand pushes. An ignore rule cannot untrack anything, so onboarding also reports which of those files are already committed and tells you togit rm --cachedthem and rotate anything exposed.
After setup
Each tool needs a nudge before it sees the server: restart PyCharm, reload the VS Code window, start a new copilot session, confirm the server is green under Cursor Settings > MCP, press Refresh in Cascade's MCP panel, approve the project MCP server in Claude Code, or trust the workspace in Codex. Setup prints the hint for the tools you selected. See Setup and dashboard and Agents and prompts next.
Was this helpful?
Your feedback shapes what we write next.
