Sync MCP server configurations and instruction files (CLAUDE.md) from Claude Code to Gemini CLI, Codex CLI, OpenCode, Kiro CLI, and Cursor.
README translations: 🇹🇼 繁體中文 | 🇨🇳 简体中文 | 🇯🇵 日本語 | 🇰🇷 한국어
If you use Claude Code as your primary AI coding agent but also switch between other agents (Gemini CLI, Codex CLI, OpenCode, Kiro, Cursor) to take advantage of their free tiers or different models, you know the pain — every tool has its own MCP config format, and setting them up one by one is tedious. Same goes for instruction files — CLAUDE.md, GEMINI.md, AGENTS.md all need the same content but in different formats.
This tool lets you configure MCP servers and write instructions once in Claude Code, then sync everywhere with a single command.
Use directly inside Claude Code with slash commands:
# Load the plugin for this session
claude --plugin-dir /path/to/sync-agents-settings
# Then use slash commands in the conversation:
# /sync-list — list all MCP servers
# /sync — sync MCP configs (with dry-run preview)
# /sync-diff — compare configs between agents
# /sync-instructions — sync CLAUDE.md to other agentsThe plugin also includes a sync-awareness skill that automatically suggests syncing when you edit MCP settings or CLAUDE.md files.
No installation needed — just run with npx:
# List all MCP servers detected from Claude Code
npx sync-agents-settings list
# Preview sync (no files modified)
npx sync-agents-settings sync --dry-run
# Sync to all targets (with automatic backup)
npx sync-agents-settings sync
# Sync CLAUDE.md instructions to all targets
npx sync-agents-settings sync-instructions# Global install for the `sync-agents` command
npm install -g sync-agents-settings
# Then use directly
sync-agents list
sync-agents sync# Sync to a specific target
sync-agents sync --target gemini
sync-agents sync --target codex
sync-agents sync --target opencode
sync-agents sync --target kiro
sync-agents sync --target cursor
# Sync to Codex project-level config
sync-agents sync --target codex --codex-home ./my-project/.codex
# Compare differences
sync-agents diff
# Skip OAuth-only servers (e.g. Slack)
sync-agents sync --skip-oauth
# Skip backup
sync-agents sync --no-backup
# Verbose output
sync-agents sync -v
# Sync instruction files (CLAUDE.md → GEMINI.md / AGENTS.md / Kiro steering / Cursor rules)
sync-agents sync-instructions
# Sync only global instructions
sync-agents sync-instructions --global
# Sync only project-level instructions
sync-agents sync-instructions --local
# Sync to specific targets
sync-agents sync-instructions --target gemini codex
# Auto-overwrite without prompts (for CI)
sync-agents sync-instructions --on-conflict overwrite
# Preview instruction sync
sync-agents sync-instructions --dry-rungit clone https://github.com/Leoyang183/sync-agents-settings.git
cd sync-agents-settings
pnpm install
pnpm dev list # Run from source
pnpm test # Run testsClaude Code is the single source of truth for MCP settings, synced to all supported targets.
┌─→ Gemini Writer ─→ ~/.gemini/settings.json
├─→ Codex Writer ─→ ~/.codex/config.toml
~/.claude.json ─────┐ │
├─→ Reader ─→ UnifiedMcpServer[] ─┼─→ OpenCode Writer ─→ ~/.config/opencode/opencode.json
~/.claude/plugins/ ──┘ │
├─→ Kiro Writer ─→ ~/.kiro/settings/mcp.json
└─→ Cursor Writer ─→ ~/.cursor/mcp.json
| Stage | Description |
|---|---|
| Reader | Reads from ~/.claude.json and enabled plugin .mcp.json files, merges into a unified format |
| Gemini Writer | JSON → JSON, type: "http" → httpUrl, ${VAR} → $VAR |
| Codex Writer | JSON → TOML, ${VAR:-default} → expanded to actual value (env value or fallback) |
| OpenCode Writer | JSON → JSON, command+args → merged command array, env → environment, type: "local"/"remote" |
| Kiro Writer | Same format as Claude, ${VAR:-default} → expanded |
| Cursor Writer | Same format as Claude, ${VAR:-default} → expanded |
Syncs CLAUDE.md instruction files to each target's native format:
┌─→ ~/.gemini/GEMINI.md (plain copy)
├─→ ~/.codex/AGENTS.md (plain copy)
~/.claude/CLAUDE.md ─→ filter @imports ──┼─→ ~/.config/opencode/AGENTS.md (plain copy)
├─→ ~/.kiro/steering/claude-instructions.md (+ inclusion: always)
└─→ ⚠ Cursor global not supported (SQLite)
┌─→ ./GEMINI.md (plain copy)
├─→ ./AGENTS.md (Codex + OpenCode share)
./CLAUDE.md ──────────→ filter @imports ──┼─→ .kiro/steering/claude-instructions.md (+ inclusion: always)
└─→ .cursor/rules/claude-instructions.mdc (+ alwaysApply: true)
| Target | Global | Local | Format Transform |
|---|---|---|---|
| Gemini | ~/.gemini/GEMINI.md |
./GEMINI.md |
Plain copy (filter @import lines) |
| Codex | ~/.codex/AGENTS.md |
./AGENTS.md |
Plain copy (filter @import lines) |
| OpenCode | ~/.config/opencode/AGENTS.md |
./AGENTS.md (shared with Codex) |
Plain copy (filter @import lines) |
| Kiro | ~/.kiro/steering/claude-instructions.md |
.kiro/steering/claude-instructions.md |
Add inclusion: always frontmatter |
| Cursor | Not supported (SQLite) | .cursor/rules/claude-instructions.mdc |
Add alwaysApply: true frontmatter |
When a target file already exists, you'll be prompted to choose: overwrite, append (keep existing + add CLAUDE.md below), or skip. Use --on-conflict overwrite|append|skip for non-interactive mode.
Safety mechanisms:
- Existing servers are never overwritten (idempotent, safe to re-run)
- Automatic backup to
~/.sync-agents-backup/by default (--no-backupto skip) --dry-runpreviews changes without writing any files
Reads MCP servers from two sources:
~/.claude.json→mcpServersobject (user-configured servers)~/.claude/plugins/cache/<marketplace>/<plugin>/<version>/.mcp.json→ enabled plugin MCP servers (matched against~/.claude/settings.jsonenabledPlugins)
Claude Code has two .mcp.json formats:
Writes to ~/.gemini/settings.json → mcpServers object.
Key format differences from Claude:
- Claude
type: "http"→ GeminihttpUrl - Claude
type: "sse"→ Geminiurl - Claude
command(stdio) → Geminicommand(same) - Env var syntax: Claude
${VAR}→ Gemini$VAR(auto-converted)
// Gemini settings.json
{
"theme": "Dracula", // existing settings preserved
"mcpServers": {
"context7": { // stdio server
"command": "npx",
"args": ["-y", "@upstash/context7-mcp"]
},
"sentry": { // http server
"httpUrl": "https://mcp.sentry.dev/mcp"
}
}
}Writes to ~/.codex/config.toml (global) by default. Use --codex-home <path> to write to a project-level .codex/config.toml instead.
Note: Codex CLI does NOT merge global and project configs. When a project has
.codex/, Codex only reads that directory. Global~/.codex/is ignored entirely.
Key format differences:
- Uses TOML instead of JSON
command/argsfor stdio (same concept)urlfor HTTP servers (no type field needed)envis a TOML sub-table[mcp_servers.<name>.env]
[mcp_servers.context7]
command = "npx"
args = ["-y", "@upstash/context7-mcp"]
[mcp_servers.sentry]
url = "https://mcp.sentry.dev/mcp"
[mcp_servers.n8n-mcp]
command = "npx"
args = ["n8n-mcp"]
[mcp_servers.n8n-mcp.env]
N8N_API_KEY = "your-key"
N8N_API_URL = "https://your-n8n.example.com"Writes to ~/.config/opencode/opencode.json → mcp object.
Key format differences:
- Root key is
mcp(notmcpServers) - stdio servers use
type: "local"with a mergedcommandarray (command + args combined) - HTTP/SSE servers use
type: "remote" - Environment variables use
environmentfield (notenv)
// opencode.json
{
"model": "anthropic/claude-sonnet-4-5", // existing settings preserved
"mcp": {
"context7": { // stdio → local
"type": "local",
"command": ["npx", "-y", "@upstash/context7-mcp"]
},
"sentry": { // http → remote
"type": "remote",
"url": "https://mcp.sentry.dev/mcp"
},
"n8n-mcp": { // env → environment
"type": "local",
"command": ["npx", "n8n-mcp"],
"environment": {
"N8N_API_KEY": "your-key"
}
}
}
}Writes to ~/.kiro/settings/mcp.json → mcpServers object.
Same format as Claude Code. ${VAR:-default} syntax in URLs is auto-expanded during sync.
Writes to ~/.cursor/mcp.json → mcpServers object.
Same format as Claude Code. ${VAR:-default} syntax in URLs is auto-expanded during sync.
| Claude Code | Gemini CLI | Codex CLI | OpenCode | Kiro CLI | Cursor |
|---|---|---|---|---|---|
command + args (stdio) |
command + args |
command + args |
type: "local", command: [cmd, ...args] |
same as Claude | same as Claude |
type: "http" + url |
httpUrl |
url |
type: "remote", url |
same as Claude | same as Claude |
type: "sse" + url |
url |
url |
type: "remote", url |
same as Claude | same as Claude |
env |
env |
env |
environment |
env |
env |
oauth |
skipped | skipped | skipped | skipped | skipped |
Every sync automatically backs up all affected config files to ~/.sync-agents-backup/<timestamp>/ before writing, preserving the original directory structure relative to ~:
~/.sync-agents-backup/2026-03-20T00-06-08-042Z/
├── .claude.json # ← ~/.claude.json
├── .claude/
│ └── settings.json # ← ~/.claude/settings.json
├── .gemini/
│ └── settings.json # ← ~/.gemini/settings.json
├── .codex/
│ └── config.toml # ← ~/.codex/config.toml
├── .config/
│ └── opencode/
│ └── opencode.json # ← ~/.config/opencode/opencode.json
├── .kiro/
│ └── settings/
│ └── mcp.json # ← ~/.kiro/settings/mcp.json
└── .cursor/
└── mcp.json # ← ~/.cursor/mcp.json
Use --no-backup to skip. Target directories that don't exist (CLI not installed) will be skipped with a warning, not created.
| Tool | Config Path | Format |
|---|---|---|
| Claude Code (user MCP) | ~/.claude.json |
JSON |
| Claude Code (settings) | ~/.claude/settings.json |
JSON |
| Claude Code (plugin MCP) | ~/.claude/plugins/cache/.../.mcp.json |
JSON |
| Gemini CLI | ~/.gemini/settings.json |
JSON |
| Codex CLI (global) | ~/.codex/config.toml |
TOML |
| Codex CLI (project) | .codex/config.toml (use --codex-home) |
TOML |
| OpenCode (global) | ~/.config/opencode/opencode.json |
JSON |
| OpenCode (project) | opencode.json in project root |
JSON |
| Kiro CLI (global) | ~/.kiro/settings/mcp.json |
JSON |
| Kiro CLI (project) | .kiro/settings/mcp.json in project root |
JSON |
| Cursor (global) | ~/.cursor/mcp.json |
JSON |
| Cursor (project) | .cursor/mcp.json in project root |
JSON |
| Tool | Global Path | Project Path | Format |
|---|---|---|---|
| Claude Code | ~/.claude/CLAUDE.md |
./CLAUDE.md |
Markdown |
| Gemini CLI | ~/.gemini/GEMINI.md |
./GEMINI.md |
Markdown |
| Codex CLI | ~/.codex/AGENTS.md |
./AGENTS.md |
Markdown |
| OpenCode | ~/.config/opencode/AGENTS.md |
./AGENTS.md |
Markdown |
| Kiro CLI | ~/.kiro/steering/claude-instructions.md |
.kiro/steering/claude-instructions.md |
Markdown + frontmatter |
| Cursor | Not supported (SQLite) | .cursor/rules/claude-instructions.mdc |
MDC (Markdown + frontmatter) |
This project can be used as a Claude Code plugin, providing slash commands and a contextual skill directly inside Claude Code conversations.
| Command | Description |
|---|---|
/sync |
Sync MCP server configs to other agents (with dry-run preview and confirmation) |
/sync-list |
List all MCP servers configured in Claude Code |
/sync-diff |
Compare MCP configs between Claude and other agents |
/sync-instructions |
Sync CLAUDE.md instruction files to other agent formats |
The plugin includes a skill that automatically detects when you're editing MCP settings (.claude.json, .mcp.json) or CLAUDE.md files, and suggests syncing to other agents.
# Validate plugin structure
claude plugins validate /path/to/sync-agents-settings
# Test locally (loads plugin for this session only)
claude --plugin-dir /path/to/sync-agents-settings- OAuth servers (e.g. Slack with
oauth.clientId) are synced as URL-only — you'll need to authenticate manually in each CLI ${CLAUDE_PLUGIN_ROOT}env vars won't resolve in other CLIs- Codex CLI doesn't support
${VAR:-default}syntax in URLs — these are auto-expanded during sync (env value if set, otherwise the default) - Re-running sync will not overwrite existing entries (safe to run multiple times)
- Codex CLI does NOT merge global and project configs — when
.codex/exists in a project, global~/.codex/is ignored - If target config directories don't exist, sync will skip that target (won't create directories)
MIT