Add your own MCP server, API or app

Add an MCP server

1. Choose Add → Add custom connector.
2. Give it a Name, which is shown in the integrations list.
3. Paste the MCP server URL, the https address where the server accepts MCP requests, for example https://mcp.example.com/mcp.
4. Optionally, add an Icon (PNG or JPG, up to 1 MB), a Note for Omne AI on how and when to use it, and Custom headers for a server that takes a key in a header.
5. Press Continue.

Omne runs three checks before anything is saved, and each row turns as soon as its check ends:

• Connecting to the server: it asks the address for an MCP answer.
• Looking up sign-in settings: it reads where the server signs people in.
• Checking the sign-in provider: it checks whether Omne can register itself there, or uses your own OAuth client.

If nothing answers, Omne says No server answered at this address. Check it for typos. You can go back, or press Continue anyway and connect it later from its card.

A server that signs in shows Sign in to its address. Omne's connect screen opens, then the server's own sign-in. Once you approve, Omne lists the server's tools, so you see what Omne AI will be able to do before you save.

Then press Save to keep it to yourself, or Save and share to give it to everyone in the workspace. An owner can also share it to the client workspaces they run, from the arrow beside Save and share.

Only use connectors from developers you trust. Omne doesn't control which tools a server offers, or promise they won't change.

Advanced settings

Open Advanced settings (optional) to give your own OAuth client ID and OAuth client secret. You need them only when the server doesn't let apps register themselves, or when you want sign-ins to use your own client.

What Omne AI does with it

Omne AI and your agents get every tool the server lists, signed in as you. A tool that only reads runs at once; anything that changes something asks you first. Your note for Omne AI goes with each tool, so it reads it before choosing one.

A connector you saved for yourself stays yours: nobody else in the workspace gets its tools.

Header values, client secrets and API secrets are sealed when you save them, and Omne never shows them again. Only their names are listed.

Add your own app

Choose Add → Add your own app. Owners and admins can add one.

1. Make the app on the platform first.
2. Pick the Platform. Omne shows what to set up in your app: the exact Return address to allow, https://app.omne.so/api/oauth/callback/<platform>, and each scope Omne asks for. Each copies in one click.
3. Give it a Name, your brand as the consent screen shows it, and paste its Client ID and Client secret.
4. Press Check and save.

Omne checks the keys with the platform first only where the platform tells a wrong secret from a right one, such as Pinterest and Reddit. A secret the platform refuses is never saved. Every other platform saves it as Unchecked, since the platform only says whether it's right when someone connects.

From then on, Connect for that platform signs in through your app instead of Omne's. Omne's connect screen shows your app's name and your workspace's logo, and the platform's consent screen shows your brand.

Your apps, under Yours, lists each one with its platform, status, client ID, permissions, app id and how many accounts run on it. Its menu checks the keys again or removes the app. An app with accounts on it is removed only together with them.

Add a custom API

Choose Add → Custom API.

1. Give it a Name and, optionally, an Icon and a Note on how and when Omne AI should use it.
2. Add its Secrets as name and value rows, such as SOME_UNIQUE_KEY_NAME.
3. Paste its Base URL, for example https://api.example.com/v1.
4. Under Test, enter a path and press Test. Omne calls the API with your secrets and shows its real answer.
5. Press Save, or Save and share.

Omne AI and your agents get two tools for it: one that reads (GET, at once) and one that sends (POST, PUT, PATCH or DELETE, after your yes). Omne AI never sees a secret's value. It writes {{NAME}} where a secret goes, and Omne fills it in. A secret it doesn't name is sent as a header of its own name, such as Authorization or X-Api-Key. Every call stays under the base address.

Import by JSON

Choose Add → Import by JSON and paste a configuration from Claude Desktop, Cursor or a server's own docs:

json { "mcpServers": { "docs": { "url": "https://mcp.example.com/mcp", "headers": { "X-Api-Key": "…" } }, "files": { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/me/Documents"] } } } 

Press Read servers. Omne lists every server before adding anything. A server at an address goes through the same checks and sign-in as Add custom connector, with its name and headers filled in. A command (STDIO) server runs as a program on a computer, so Omne says so and doesn't import it: Omne runs on the web, and its desktop app doesn't host MCP servers yet.

The checked mark

Every integration Omne built or checked shows a thin grey mark after its name. Hover it to see who made it: Omne, or the company whose official MCP server Omne checked. What you add yourself never carries the mark. Its card shows its icon, its name, the server's address and a Custom label instead.

In the API and MCP

The same actions are tools for your own agents. With a key from Settings → MCP & API:

• GET /v1/connectors/list, /v1/connectors/get, /v1/connectors/tools, and POST /v1/connectors/check and /v1/connectors/test-api read and check.
• POST /v1/connectors/add, /v1/connectors/update and /v1/connectors/remove change them, with a key made to write.
• Your own apps use Zernio's addresses: GET and POST /v1/connect/oauth-apps, GET, PATCH and DELETE /v1/connect/oauth-apps/{appId}, and POST /v1/connect/oauth-apps/{appId}/test.

Omne's MCP server has the same tools under the same names, such as connectors_add and oauth_apps_list.