You update your MCP server. You open Claude Desktop and edit the config. Then you open Cursor and do it again. Then Windsurf. Then VS Code with Continue. Then Groq Desktop. By the time you finish, you have five JSON files on your machine that are supposed to be identical but never quite are. One has an old path. One has a typo you introduced at 11pm. One is still pointing at the version you deprecated two weeks ago.
This is the MCP multi-app config problem, and it is not a small annoyance. Every developer running more than one AI app on their machine is managing this by hand, right now, today.
"The config file is not the source of truth. You are. And humans are terrible version control systems."
MCP config files are app-local. Each AI application — Claude Desktop, Cursor, Windsurf, VS Code, Continue, Groq Desktop — maintains its own JSON file in its own directory. There is no shared config layer. There is no sync mechanism. There is no single source of truth.
When you install a new MCP server, you edit one config. When you update a server path, you edit one config. When you rotate a credential, you edit one config. Except you do it five times, in five different directories, with five different chances to introduce drift.
vek-sync pushVek-Sync is a small CLI utility bundled inside VEKTOR Slipstream. It maintains a canonical mcp-sync.json in your VEKTOR config directory. This file contains your MCP server definitions exactly once. When you run vek-sync push, it writes the correct entries into every detected AI app config on your machine.
It does not overwrite your entire config file. It performs a surgical merge — your existing app-specific settings are preserved, the MCP server block is updated. If an app is not installed, that target is skipped silently. If a config file does not exist yet, it is created.
// Your single source of truth
{
"mcpServers": {
"vektor-slipstream": {
"command": "node",
"args": ["/home/user/.vektor/vektor.mjs", "mcp"],
"env": {
"VEKTOR_LICENCE_KEY": "vk-xxxx-xxxx"
}
}
}
}# Push to all detected apps
$ vek-sync push
# Output
✓ Claude Desktop → /Users/you/Library/Application Support/Claude/claude_desktop_config.json
✓ Cursor → /Users/you/.cursor/mcp.json
✓ Windsurf → /Users/you/.codeium/windsurf/mcp_config.json
✓ VS Code → /Users/you/.vscode/settings.json
✓ Groq Desktop → /Users/you/Library/Application Support/Groq/mcp.json
- Continue → not installed, skipped
Done. 5 apps synced in 120ms.Config drift is not just inconvenient — it is a debugging nightmare. When a tool stops working in Cursor but keeps working in Claude, the natural assumption is that something is wrong with the tool, the MCP server, or Cursor's protocol implementation. The actual answer is usually that Cursor is running the old version of the server binary because you updated the path three weeks ago and forgot to propagate it.
Vek-Sync includes a vek-sync diff command that reads all detected app configs and shows you exactly where they diverge from your canonical source. Green means in sync. Red means the app is running something different from what you think it's running.
$ vek-sync diff
✓ Claude Desktop in sync
✕ Cursor DRIFT — args[0] points to v1.2.1, canonical is v1.3.0
✓ Windsurf in sync
✕ VS Code DRIFT — VEKTOR_LICENCE_KEY missing from env block
✓ Groq Desktop in syncThis is the output that would have saved you two hours of debugging. Run vek-sync push and the drift is resolved in under a second.
One of the most painful MCP workflows is rotating a credential. You generate a new licence key, or you need to update an API token that your MCP server uses. In a no-sync world, you open every config file, find the env block, update the value, save, repeat. Miss one and the app starts failing with an auth error at 2am.
With Vek-Sync, credential rotation is a two-step operation. Update mcp-sync.json once. Run vek-sync push. Every app gets the new credential atomically. There is no partial state. There is no "I think I got them all."
Store your mcp-sync.json in a private git repository. Every vek-sync push becomes a one-line commit. You now have a full audit trail of every MCP configuration change across every app on your machine, timestamped and diffable forever.
Vek-Sync auto-detects config paths for all major AI apps on macOS, Windows, and Linux. It knows where each app stores its MCP config by default and handles the platform differences silently. On Windows it resolves %APPDATA% paths correctly — yes, including the Node.js path quoting issues that plague manual Windows configs.
Vek-Sync ships as part of VEKTOR Slipstream — no separate install, no extra dependency. Once you have VEKTOR set up, the vek-sync commands are available immediately. The canonical config lives alongside your VEKTOR memory database, so your entire agent infrastructure — memory, vault, SSH orchestration, and MCP config — is in one place.
For teams using VEKTOR across multiple machines, mcp-sync.json can be committed to a shared private repo and pulled on each machine. Run vek-sync push after a pull and every machine in the team is running identical MCP configs within seconds. No Slack messages asking "what path are you using for the server?" No onboarding docs with stale config snippets.
Vek-Sync merges into app config files — it does not replace them. App-specific settings unrelated to MCP servers are never touched. If you have custom Cursor or VS Code settings outside the MCP block, they are preserved exactly.
The MCP ecosystem is growing fast. New servers are being published daily. New AI apps are adopting the protocol every month. The more servers you run and the more apps you use, the worse the manual config management problem gets. It scales quadratically: N servers × M apps = N×M places where something can be wrong.
Vek-Sync collapses that to N. One config. One place to update. One command to propagate. The problem that currently scales with your AI tooling stack stops scaling entirely.
If you are running MCP servers today and you are not using a sync layer, you are already in config drift. You just might not have noticed yet.