Use Odyn from your AI tools
Connect Claude, ChatGPT, Cursor and other MCP clients to your Odyn projects, so your own AI assistant can read your code, edit files and deploy to staging.
Odyn works with any AI tool that supports MCP (Model Context Protocol). Connect it once and your assistant can see your projects, read and edit files, check what would change, and deploy to staging, with your own model and your own setup.
Odyn's built-in AI keeps working as before. This is for people who already live in Claude, ChatGPT, Cursor or a coding agent and want Odyn there too.
Connect a tool
The server address is:
https://mcp.odyn.dev/mcpThe first time you connect, your tool opens an Odyn sign-in page. Sign in, choose the workspace it can work in, review what it's asking for, and approve. Your assistant can then read, edit and deploy to staging in that workspace. Production deploys always wait for your approval in Odyn. To move a tool to another workspace, connect it again or change it in Settings > AI. A tool connected this way sees only that workspace's projects, not single projects you were invited to elsewhere; a personal key can reach those.
Claude Code
claude mcp add --transport http odyn https://mcp.odyn.dev/mcpThen run /mcp in Claude Code and choose Authenticate.
Claude (web and desktop)
In Claude's settings, add a custom connector and paste the server address.
ChatGPT
In ChatGPT's settings, add a custom connector (this may need developer mode) and paste the server address.
Cursor
Add this to .cursor/mcp.json (or your global Cursor MCP settings):
{
"mcpServers": {
"odyn": { "url": "https://mcp.odyn.dev/mcp" }
}
}Other tools
Any client that supports remote MCP servers with sign-in works. Tools that only run local servers can connect through mcp-remote.
Tools that can't sign in
Create a personal key in Settings > AI and choose what it can do. Your tool sends it as Authorization: Bearer <key>. Copy it when it's shown, since Odyn shows each key only once, and revoke it there when you're done.
What your assistant can do
| Area | What it can do |
|---|---|
| Projects | List your projects and see each one's build mode, plan, and last deploy errors |
| Files | List, read and search files, and see each file's version |
| Edits | Change exact text in a file, create files, replace a file, rename or move it, delete it |
| Checks | Validate JavaScript, TypeScript or CSS the way Odyn's bundler will |
| Deploys | See what would change, deploy to staging, get your embed codes |
| Production | Ask to deploy to production. You approve it in Odyn first |
Your plan's limits apply the same way they do in the editor.
Working alongside the editor
You can keep the editor open while your assistant works.
- Nothing gets lost. Edits are checked against the file's current version. If you and your assistant change the same file, Odyn shows a conflict instead of losing either change.
- No surprise deletes. Your assistant can't delete a file while someone has the project open and active in the editor.
Production needs your approval
When your assistant asks to deploy to production, it gets a link instead of deploying. Open it in Odyn to see exactly which files and libraries would change, then approve or reject it. Approving asks you to confirm it's you.
Only the files you approved are deployed. If anything changes before the deploy runs, your assistant has to ask again.
Projects whose staging runs on a client's custom domain work the same way for staging deploys.
Good to know
- Your assistant sees the files saved in Odyn, not unsaved changes in your editor.
- Your assistant can add and remove CDN libraries like GSAP, and they load in the order your dependencies panel shows.
- Anything your assistant reads from your project is treated as data, never as instructions, but review what it plans to do before you let it deploy.
Disconnect a tool
Open Settings > AI in Odyn. Your connected tools are listed there. Choose Disconnect and the tool loses access right away, even if it's still signed in. To use it again later, choose Allow again.