Tokens and authentication
Three kinds of token
| Kind | Starts with | Who creates it | What it reads | Server | Expiry |
|---|---|---|---|---|---|
| Organization | anmcp_o_ | A member of your organization with Use AI agents (MCP) | That one organization, with the creator’s permissions | https://mcp.appneck.com/ | Up to 365 days (default 90) |
| Platform | anmcp_p_ | Appneck staff with two-factor authentication on | Every organization, read-only, for Appneck’s own support and operations | https://mcp.appneck.com/platform | Up to 90 days (default 30) |
| Service | anmcp_s_ | Appneck staff, for an internal Appneck system | Only the organizations on its allowlist, and only those with AI agents switched on | https://mcp.appneck.com/ | Up to 365 days |
Most of this page is about organization tokens, the ones you create. The other two kinds belong to Appneck; they are described at the end so you know what can read your data and how.
A token of one kind never works on the other server: an organization or service token is refused on /platform, and a
platform token is refused on /.
Creating a token
Open AI agents in the Appneck panel. You need the Use AI agents (MCP) permission (Owners, Admins and Developers have it by default) and a plan that includes AI agents.
A token has:
- A name, e.g. “Claude Code on my laptop”, so you can tell tokens apart later.
- Scopes: what it may read. You can only pick scopes your own role allows. See the scopes reference.
- An expiry: 30, 60, 90 (default), 180 or 365 days. Expiry is mandatory, and it can be shortened but never extended. Create a new token instead.
The token looks like anmcp_o_ followed by 46 letters and digits. It is shown once; Appneck stores only a hash
of it. If you lose it, revoke it and create a new one.
Using a token
Send it in the HTTP Authorization header, never in a URL:
Authorization: Bearer anmcp_o_…Your panel login and your MCP tokens are separate: a panel session never works on the MCP server, and an MCP token never works in the panel or the Appneck API.
What a token can do, call by call
On every call, Appneck takes the token’s scopes and keeps only the ones the creator’s current role still allows. If an Owner removes a permission from your role, tokens you created lose the matching scope on their next call. The token itself keeps its scopes: give the permission back and they work again.
When a token stops working
| Event | What happens |
|---|---|
| The token expires | Calls are refused. You get an email 7 days before. |
| You or an admin revoke it | Calls are refused at once. If an admin revoked yours, you get an email. |
| You leave the organization, or your account is disabled | All your tokens are revoked automatically. |
| Your role loses the Use AI agents (MCP) permission | Calls are refused at once, and within the hour your tokens are revoked (you get an email). Getting the permission back doesn’t revive them: create new ones. |
| The organization is suspended | Calls are refused until it is reactivated; tokens are kept. |
| The plan no longer includes AI agents | Calls are refused; tokens are kept. |
Owners and Admins (the Manage AI agent access permission) see every token in the organization and can revoke any of them.
Platform tokens (Appneck staff)
Appneck staff use platform tokens to answer support and operations questions across organizations. They are read-only and shown once. Each one belongs to one staff member, and three rules apply:
- Two-factor authentication, when Appneck requires it (the default): a staff member needs it on to create a platform token, and it is checked again on every call — if it is turned off, calls are refused at once and the token is revoked within the hour.
- Permissions are checked on every call. A staff member who loses the platform AI agents permission, or whose account is disabled, is refused at once, and their tokens are revoked.
- Personal data is masked. Emails and names are masked unless the token has
pii:read, and searching by email or name needs it too.
Every new platform token is announced by email to the staff who oversee MCP access. Every call is logged with the organization it read.
Service tokens (Appneck systems)
A service token lets an internal Appneck system, such as Rankevio’s agents, read organizations through the same tools as your own tokens. Its allowlist is either a fixed list of organizations or “organizations where a given feature is on”, never all organizations. On every call it must name one organization:
{ "name": "list_products", "arguments": { "organization_id": "01a1c9d0-…" } }The call works only when that organization is on the allowlist and has AI agents (MCP) switched on. Otherwise the answer is “not found”, the same as for anything outside a token’s reach. Service tokens follow the same personal-data masking as your own tokens. Their calls appear in your organization’s usage with the organization they read.
Connecting by signing in (OAuth)
Sign-in connections are switched on separately from the rest of AI agents. Until they are, your AI agents page has no Connected apps tab and Appneck answers “not found” to the steps below; while they are being rolled out, the Appneck page may say “Not available yet”. In both cases, connect with a token instead.
Claude.ai, ChatGPT, Claude Desktop’s connectors and Claude Code can connect by signing in with Appneck instead
of a pasted token. You add the server address https://mcp.appneck.com/ to the app; it opens an Appneck page where
you choose what it may read. Guides: Claude.ai, ChatGPT,
Claude Code.
What you approve
The Appneck page at appneck.com/connect/authorize shows:
- The app’s name and web address. Apps that publish a client metadata document (Claude.ai, Claude Code and others) are identified by that address. An app that registered itself on the spot is marked Unverified app: Appneck can’t confirm who runs it, and the name is what the app says about itself. Connect it only if you started the connection from an app you trust.
- One organization. Only organizations where your role has the AI agents (MCP) permission and the plan includes AI agents are listed. The app sees that organization only.
- What it may read. Only areas your role allows are offered. Personal data is off unless you tick it.
You sign in with your normal Appneck login, with two-factor authentication if you use it. Deny sends you back to the app with nothing shared.
While it is connected
- The app reads with the access you approved, limited to what your role allows at the time of each call. If your role loses a permission, the app loses that area at once. If you leave the organization, your account is disabled, or the plan stops including AI agents, the app’s next call is refused.
- Read only, like tokens: the app can never change anything.
- Each connection has its own limit of 60 calls a minute, inside your organization’s limit.
- The app’s access lasts an hour at a time and is renewed in the background for up to 30 days of inactivity. Each renewal replaces the previous one. If an old renewal token is ever used again, Appneck treats it as stolen, disconnects the app and emails you.
- Connecting the same app again replaces the earlier connection, including its organization.
Disconnecting
AI agents → Connected apps lists your connections (an Owner or Admin with AI agents management sees everyone’s). Disconnect ends the connection at once: the app’s next call is refused and it can’t renew. You get an email when someone else disconnects your app, when Appneck disconnects it after a reused renewal token, or when a permission change ends it.
Technical details (for client developers)
| Authorization server metadata | https://appneck.com/.well-known/oauth-authorization-server (RFC 8414) |
| Protected resource metadata | https://mcp.appneck.com/.well-known/oauth-protected-resource (RFC 9728), also in the 401’s WWW-Authenticate header |
| Resource / audience | https://mcp.appneck.com — send it as resource (RFC 8707); tokens for any other audience are refused |
| Grants | authorization_code with PKCE S256 (required), refresh_token (rotating). No client secrets |
| Client registration | Client ID Metadata Documents (an https URL as client_id) preferred; Dynamic Client Registration at /oauth/register |
| Redirect URIs | Exact match; loopback (http://127.0.0.1 or http://localhost) may use any port (RFC 8252) |
| Authorization response | Includes iss (RFC 9207) |
| OAuth scope | mcp:use. What the app may read is chosen on the consent page, not in the scope parameter |
Client metadata documents must be served over https without redirects, under 5 KB, within 3 seconds, and their
client_id must equal the document’s URL exactly.