Agent Cockpit

Agent Cockpit

macOS desktop control panel for AI coding agent configuration. See and edit every **MCP server, skill, subagent, slash command, plugin, settings file and instruction doc** across **Claude Code, OpenAI Codex and Cursor** — in one window, with diff previews and automatic backups before every write.

Why

Agent configuration is scattered across many files in different formats:

Surface	Claude Code	Codex	Cursor
MCP servers (user)	~/.claude.json	~/.codex/config.toml	~/.cursor/mcp.json
MCP servers (project)	~/.claude.json + .mcp.json	—	.cursor/mcp.json
Skills	~/.claude/skills/	~/.codex/skills/	~/.cursor/skills/
Subagents	~/.claude/agents/	—	~/.cursor/agents/
Slash commands	~/.claude/commands/	—	—
Plugins	settings.json + installed_plugins.json	config.toml	—
Instructions	CLAUDE.md / AGENTS.md	AGENTS.md / rules	—

Agent Cockpit reads all of them into one inventory and writes them back **surgically**.

Safety model

Editing these files by hand (or with a naive tool) is risky — `~/.claude.json` holds ~70 unrelated state keys, and `config.toml` has comments you don't want to lose. Every write in Agent Cockpit goes through the same pipeline:

1. Surgical edits only — JSON is edited via jsonc-parser text edits (only the target subtree's bytes change; never a whole-file re-serialize). TOML is spliced via AST ranges (toml-eslint-parser), so comments and formatting outside the touched block survive byte-for-byte.
2. Diff preview — every save shows the exact file diff first. Env var / header values are masked in diffs.
3. Conflict detection — the file's hash is checked at apply time; if Claude Code or your editor changed it since preview, you get a conflict banner instead of a silent overwrite.
4. Automatic backups