Ikuna MCP

Connect your AI client to the workspace context in the Ikuna app running on your Mac.

ikuna-mcp 0.1.1 is listed in the official MCP Registry as io.github.RobertoHanas/ikuna-mcp.

Connect

1. Install Node.js 18 or later, including npx.
2. Open Ikuna and enable the capabilities you want to share in Settings → AI Connections.
3. Run the guided setup:

npx -y ikuna-mcp setup

Setup detects Claude Desktop, Codex, Claude Code, and Cursor, merges only Ikuna's entry into their existing configuration, retains the latest three backups, and verifies the local MCP handshake. Re-run the same command to repair stale or missing values, or make that intent explicit:

npx -y ikuna-mcp setup --repair
npx -y ikuna-mcp setup --client codex

Exit status 0 means every detected client was configured and the handshake passed, 1 means configuration was saved but verification needs a manual step, and 2 means no detected client could be configured.

For a client that is not detected automatically, add this entry to its mcpServers configuration, keeping your other servers:

"ikuna": { "command": "npx", "args": ["-y", "ikuna-mcp"] }

For clients such as Claude Desktop, Claude Code, and Cursor, a complete JSON configuration is:

{
  "mcpServers": {
    "ikuna": {
      "command": "npx",
      "args": ["-y", "ikuna-mcp"]
    }
  }
}

For Codex, add this to ~/.codex/config.toml:

[mcp_servers.ikuna]
command = "npx"
args = ["-y", "ikuna-mcp"]
startup_timeout_sec = 30

Restart or reconnect your MCP client after saving. Ask it to check Ikuna's health, then ask about your workspace. Your available tools depend on the capabilities enabled in Ikuna.

The first run downloads this small package from npm. Later runs use npm's cache. If your client cannot find npx, use the full executable path reported by command -v npx as command and ensure Node.js is on that client's PATH.

Requirements and troubleshooting

This package runs locally on macOS. It requires an Ikuna build that includes the MCP helper at Contents/bin/ikuna-mcp.

• Ikuna is installed but closed: open Ikuna, leave it running, and reconnect the MCP client. The npm bridge exits with an error when Ikuna is closed; it does not open the app for you.
• Ikuna is not installed: install it from the official Ikuna product page, open it, and reconnect the MCP client.
• MCP helper unavailable: update Ikuna to a build with AI Connections support.
• Discovery timeout: quit and reopen Ikuna, then retry. Running-app discovery has a short deadline and errors are written to stderr, never mixed with MCP responses.
• More than one Ikuna copy running: quit the extra copies. All copies share the local bridge service, so use one running installation at a time.
• Tools unavailable or approval required: review the capabilities in Ikuna's AI Connections settings. Installing this package does not change permissions.

Custom installation locations are discovered automatically from the running app. To select a particular running bundle explicitly:

"ikuna": {
  "command": "npx",
  "args": ["-y", "ikuna-mcp", "--app-path", "/Applications/Ikuna.app"]
}

IKUNA_APP_PATH is the equivalent environment variable. A selected bundle must already be running. The optional IKUNA_MCP_CLIENT_ID environment variable is passed through to Ikuna for connection bookkeeping; it does not grant access.

How it works

MCP client → npx ikuna-mcp → Ikuna.app/Contents/bin/ikuna-mcp → local XPC bridge → Ikuna

The npm package discovers the running Ikuna bundle and launches its embedded native helper with the client's stdio streams. The native helper provides the signed XPC connection, request deadlines, and per-client protocol sessions. Ikuna remains responsible for tools, workspace data, approvals, and mutations. On the way back to the client, the package repairs the tools/list schema so an optional confirmation field stays optional and strict clients do not reject valid read calls; it changes no request, workspace data, or result. The package has no runtime dependencies, reads no workspace database, and opens no network listener.

The MCP Registry manifest is included as server.json. Maintainers can follow RELEASE.md to publish the npm package and listing together.