Omniscio

Sync to Other AI Tools

Sync to Other AI Tools keeps the setup you manage inside Omniscio available in the other AI coding tools on your computer: Codex, Gemini, OpenCode, Cursor, Grok, Antigravity, Pi and Hermes. It is on by default for your skills and it targets every tool you have enabled, so most people never need to touch it. Your instructions and MCP servers are only shared if you opt in, and a server that carries a password or key is never copied anywhere.

On this page

What it is

Sync to Other AI Tools takes the setup you already manage inside Omniscio - your skills, your agent instructions, and your custom MCP servers - and installs it into the other AI coding tools on your machine: Codex, Gemini, OpenCode, Cursor, Grok, Antigravity, Pi and Hermes. You set a thing up once in Omniscio and it follows you into every tool, even when you use those tools outside Omniscio.

What is on and what is off

Part Default
The master switch On
Sync skills On
Sync instructions Off
MCP servers Off - you pick each server
Background re-sync every 6 hours Off
Copy skills made in other tools back into Omniscio Off

The master switch and skills sync are on for new installs and for existing ones alike. Installs that existed before this release were brought onto the new default once; if you switch it off afterwards, it stays off. Instructions, MCP servers, the background timer and the copy-back option are always your choice.

It aims itself at the tools you actually use: every AI tool you have enabled in Omniscio is a sync target by default, so there is nothing to tick before it does something useful. A tool you have not turned on receives nothing, and a tool that is not installed on the machine is listed as "not installed" and left alone.

For skills this means real portability: Omniscio copies each skill's whole folder (its scripts and reference files too) into the other tool's skills directory, so the skill is actually runnable there, not just a list of names. The copies stay live: with skills sync on, editing a skill in ~/.claude/skills gets re-synced to the other tools by a light background pass the next time you start a session (throttled, never a constant file watcher). They are real copies, not fragile links, and they are edit-protected: if you hand-edit a synced copy inside another tool, Omniscio leaves it alone and flags it rather than overwriting or deleting your change.

This is the opposite direction from Omniscio's normal MCP behavior. Normally Omniscio delivers MCP servers per session, writing a private per-session .mcp.json (under Omniscio's data folder, never your project folder) only for the Claude sessions Omniscio itself starts, and it deliberately never touches a tool's global config (see mcp-servers.md). This feature is a separate machine-sync path that pushes a flagged subset of the same server list out to each installed tool's own global config. The per-session path is left completely untouched.

Where to find it

Open Settings → Automations & Rules → Sync to Other AI Tools. The page holds the master switch, the target tools, the per-server sync toggles, the instructions and skills switches, the Cursor project-folder picker, the background auto-sync switch, the per-tool skill budgets, and the Sync now button. Typing "sync", "mcp sync" or a tool's name into the Settings search box also finds it.

It is a desktop-only page: it installs things into tools on this computer, so it is hidden on the phone view. The phone cannot start a sync, turn the feature on, or change where it writes.

How it behaves

How to use it

  1. Check the master switch. It is on by default. Switch it off and nothing is written to any other tool.
  2. Target tools - automatic by default. The switch "Sync to every AI tool you have enabled" is on, so the tools you actually use are the tools that get set up. Turn it off to tick the tools yourself - that is also how you hold a single tool back without turning it off everywhere else. Anything you tick while the switch is on is simply added to the tools you already have enabled.
  3. Choose which MCP servers to sync. Each of your custom MCP servers has a "sync to machine" toggle, off until you turn it on. A server that carries a credential cannot be synced; it is shown disabled with a lock and a "Contains a secret - not synced" note (see Safety rules below).
  4. Optionally turn on Sync instructions (mirror your own agent instructions into each tool's global instructions file) and, if you target Cursor, add the project folders you want your skills installed into (each gets a .cursor/skills/ folder; Cursor has no machine-wide skills folder).
  5. Optionally turn on Auto-sync in the background to re-sync about every 6 hours.
  6. Click Sync now to run a full sync. You get a per-tool result: servers added, already present, not added (with the reason) or failed; the instructions status; and a skills tally (copied, unchanged, an amber edited count for copies you changed in the other tool, removed for copies pruned because their skill is gone, skipped, failed, or already available for OpenCode and Grok).

How skills reach each tool

  • Codex, Gemini, Antigravity, Pi and Hermes read skills from their own folder, so Omniscio copies them there. Until Omniscio has actually copied skills into a tool's folder during the current app session, sessions in that tool are given a compact list of your skills instead, so a brand-new session is never left with no skills. After a successful copy they find the real folders.
  • OpenCode and Grok read your ~/.claude/skills folder directly, so nothing is copied and their row shows already available.
  • Cursor has no machine-wide folder, so Omniscio copies the skills into the session's workspace just before a Cursor session starts, plus any project folders you picked.
  • Claude is the source, so it is never a copy target.

What gets written where

  • MCP servers go into each tool's global MCP config:
    • Claude, Codex, Gemini - through that tool's own mcp add command (user scope).
    • Cursor - merged into ~/.cursor/mcp.json.
    • OpenCode - merged into ~/.config/opencode/opencode.jsonc.
    • Hermes - merged into its config.yaml (under mcp_servers), in whatever folder Hermes uses on your platform.
    • Pi - nothing; Pi has no MCP support.
    • Grok - nothing; Grok already reads ~/.claude.json through its Claude Code compatibility layer, so pushing the servers again would list each one twice.
  • Instructions (only when Sync instructions is on) go in a clearly marked managed block in each tool's global instructions file: Codex (~/.codex/AGENTS.md), Gemini (~/.gemini/GEMINI.md), OpenCode (~/.config/opencode/AGENTS.md), Grok (~/.grok/AGENTS.md), Pi (~/.pi/agent/AGENTS.md). Your own text around the block is preserved. Hermes is deliberately left out - its only file for this is SOUL.md, its identity document - and Claude is the source, and Cursor has no global instructions file. Codex and Grok accounts that Omniscio manages run in their own private home folder, so they get the same block there (a folder that no longer exists is skipped, never recreated).
    • The block holds only your own instructions: the presets you enabled and your custom text. It never carries Omniscio's own reply-format rules, which Omniscio's sessions get on their own. With Sync instructions off, or nothing to share, the block is removed and the rest of the file is left exactly as it was. Editing your instructions in Omniscio re-runs the sync.
  • Skills go into each tool's skills directory:
    • Codex - ~/.agents/skills/. Codex also scans ~/.codex/skills/, so Omniscio removes its own earlier copies from there; only copies carrying Omniscio's marker are removed, never Codex's built-in skills or anything you installed. If one of those earlier copies was edited, it is kept and reported rather than deleted.
    • Gemini - ~/.gemini/skills/.
    • Antigravity - ~/.gemini/antigravity-cli/skills/ (the agy command-line tool's folder, not the Antigravity IDE's).
    • Pi - ~/.pi/agent/skills/. A Pi session on a local Ollama model uses a private folder instead, so synced skills show up for hosted Pi models only.
    • Hermes - its own skills folder: AppData\Local\hermes\skills on Windows, ~/.hermes/skills on macOS and Linux.
    • Cursor - a .cursor/skills/ folder inside each project folder or session workspace.
    • Antigravity, Pi and Hermes never send their own built-in skills back into your library, and Grok's built-in skills in ~/.grok/skills are never touched or pulled in.

Safety rules

  • A server with a credential is never written to another tool. Omniscio treats a server as carrying one if any of these holds a credential: its environment values, its headers, its URL (a password inside the address, or a query like ?token=), its launch command, or any launch argument (--api-key=abc, --token abc, a bare --key abc, KEY=value, a database connection string such as host=db password=x), or a credential-named field in its JSON. Detection is conservative on purpose: a false alarm only means "not synced".
  • Other tools' config files are add-only. Omniscio only ever adds a server a tool is missing. An existing entry with the same name (or the same slug) is never replaced or removed, so anything you set up yourself in that tool is safe. Omniscio reads or writes nothing at all when there is nothing to add.
  • A config file with comments is left untouched. If a tool's config file (such as opencode.jsonc or Hermes's config.yaml) contains comments, Omniscio will not rewrite it, because rewriting would lose them; the servers are reported as "not added - has comments" so you can add them by hand.
  • Careful writes. Writes follow a symlinked config to its real file, keep the file's permission bits, and make a one-time .amc-bak backup of your original before the first change.
  • Skills are copied atomically (staged, then swapped in, never half-written). Every copy carries a small .amc-synced.json marker so Omniscio only ever manages copies it made; a same-named folder you or another tool created is left alone and reported as skipped.
  • Edited copies are never deleted. A skill copy you changed inside another tool is reported as edited and kept, both on refresh and when its skill is removed from your library.
  • Removal propagates safely. When a skill leaves your library (you delete it or uninstall its plugin), Omniscio's own unedited copies are pruned from the other tools on the next sync. If Omniscio sees an empty skill list it prunes nothing, so a glitch can never wipe the other tools' skills.
  • Fail-soft. If one tool or one skill errors, the rest still sync and the failure shows on that tool's row. A tool that is slow to list its servers is skipped for that run rather than risking duplicates.

Copying skills back in (off by default)

A separate setting pulls a skill you authored in Codex, Gemini or OpenCode into your Omniscio skills folder before the next sync. Its switch, Also bring in skills from other tools, sits right under Sync skills and is off for everyone, because it writes into your main skills folder; when enabled it is copy-if-absent and never touches or removes the original in the other tool. Turning on skills sync does not turn it on.

Cost on your computer

The sync is built to cost almost nothing. The refresh that runs about every 10 minutes while you start sessions only handles skills and only copies into tools whose folder already exists; tools are synced one at a time; and one deferred skills pass runs about a minute after the app starts. A "nothing changed" check reads no skill file contents - it only looks at file names, sizes and modified times, and reuses the last result while those are unchanged. On a large real skill library this took the check from roughly 21 MB read and 8 MB written per target tool down to zero bytes. One known gap: the first Cursor session in a brand-new workspace still copies the library once.

Project context broker

Omniscio also has a small read-only project-context helper (amc-broker) used in internal builds only. It is never installed on customer builds, and its note never appears in the instructions synced to your tools.

The servers this feature syncs from, and the per-session delivery it is the opposite of, are described in mcp-servers.md; read that first if you are unsure which of the two paths you want. The sibling mirror that pushes project instructions from CLAUDE.md out to AGENTS.md is agent-instructions-sync.md, and it shares the same mirror engine. The skills being copied are the ones covered by use-skills.md. When a tool such as Codex would receive more skills than it can load, skill-budget-alert.md explains the alert and how to hold some back.

Last verified 2026-10-11