Skip to content

MCP server

Sealcord’s MCP server lets an AI assistant work with your organization. It can look up a customer’s licences and machines, issue, revoke, expire and reinstate licences, free seats, see your payment integrations and manage their Polar mappings, see your organization and manage its team, and see your webhook endpoints and their deliveries. The tools page lists each one.

The server is a client of the Admin API, so every rule, scope and limit stays the API’s. It reaches only your own organization.

The server is at https://mcp.sealcord.com/mcp. It speaks Streamable HTTP, so any client that supports remote MCP servers can connect. In Claude it is listed in the connector directory, so you connect it there without adding a custom connector.

Every client connects in one of two ways: with your sign-in, or with an admin API key. Pick yours below for the exact steps.

Client With your sign-in With an admin API key
Claude Code Yes Yes
Claude Desktop Yes, from the connector directory Where request headers are enabled (beta), or through mcp-remote
claude.ai Yes, from the connector directory Only where request headers are enabled (beta)
ChatGPT On the web, on some plans No on the web; the desktop app works like Codex
Codex CLI Yes Yes
Cursor Not yet: use an admin API key Yes
Antigravity Yes, by its docs; you paste a code back Yes
Hermes Agent Yes Yes
OpenClaw Yes Yes
Any other client If it implements MCP authorization If it can send a header, or with mcp-remote

The assistant signs in through the Console, where you choose what it may do. When your client starts the sign-in, a browser opens on the Console. Sign in as a developer with your passkey or an emailed link (a code sign-in cannot grant access), check what the assistant asks for, untick what it should not have, and choose Allow. The client keeps the tokens and refreshes them, so there is nothing to copy, unless it goes unused for 7 days: then it signs in again.

A client finds the sign-in server from the MCP server’s 401 answer and runs the OAuth authorization code flow with PKCE. It identifies itself with a Client ID Metadata Document, or registers itself, so you never create a client id.

What you may grant depends on your role. Owners and admins can grant every scope except products:write and licences:anonymise; support and read-only members can grant read scopes, and every role can grant the workspace scopes. The grant is for the organization your session acts in: the assistant sees and changes that organization’s products, licences and team, and nothing else. The workspace scopes are yours, not the organization’s: they let the assistant list your workspaces, create one you own, and move a product between two you own.

An assistant you connected before a scope existed keeps what you granted then. To grant a new scope, connect it again. In Claude Code, run /mcp, pick sealcord and authenticate again.

Actions an assistant takes with your sign-in are recorded in the licence’s audit log as you, made through the MCP server.

For scripts, or a client you configure by hand, send an admin API key as Authorization: Bearer sc_…. Give it the fewest scopes the job needs. The key’s scopes decide what the assistant can do: a read-only key gives a read-only assistant. A key without licences:write makes every licence and machine write tool fail with insufficient_scope.

Name the key after the assistant, so its actions are easy to find in the audit log and it can be revoked alone. Where a client can read the key from an environment variable, the examples below use SEALCORD_TOKEN for the key itself. Where the key has to sit in a file in plain text, use a read-only or short-lived key there, or connect with your sign-in instead.

Terminal window
claude mcp add --transport http sealcord https://mcp.sealcord.com/mcp

Then run /mcp in Claude Code, pick sealcord and choose Authenticate, or run claude mcp login sealcord. The sign-in opens in your browser.

Open Sealcord in Claude’s connector directory, or search for Sealcord in the directory, choose Connect and sign in.

You can also add it as a custom connector: open Customize, then Connectors, choose Add, then Add custom connector, name it Sealcord, enter the URL https://mcp.sealcord.com/mcp, keep the default sign-in settings and choose Continue. The Free plan allows one custom connector.

On a Team or Enterprise plan, an owner adds the connector once in Organization settings, then Connectors, with Add, then Custom, then Web. Each member then connects it under Customize, then Connectors.

Claude reaches the server from Anthropic’s cloud, not from your computer, so this works the same in Claude Desktop, claude.ai and the mobile apps.

Connect Sealcord from Claude’s connector directory and sign in, as in Claude Desktop. The same connection works in Claude Desktop and the mobile apps.

On chatgpt.com, open Plugins, choose the plus button, then Add custom MCP server. Enter the name Sealcord and the server URL https://mcp.sealcord.com/mcp, choose OAuth for authentication, confirm the warning and choose Create as a plugin. Install the plugin, sign in, then type @ in a chat to pick Sealcord.

OpenAI offers custom MCP servers on the Business, Enterprise and Edu plans: on Business only admins can use developer mode, elsewhere an admin grants access. On Pro, in developer mode, they can only read. By default, ChatGPT asks before a tool that changes something.

The Codex CLI, the Codex IDE extension and the ChatGPT desktop app share ~/.codex/config.toml, so this connects all three.

Terminal window
codex mcp add sealcord --url https://mcp.sealcord.com/mcp

It finds the sign-in and opens it in your browser. Later, codex mcp login sealcord signs in again.

Cursor keeps its servers in ~/.cursor/mcp.json, or .cursor/mcp.json for one project.

Cursor’s sign-in registers a redirect address Sealcord refuses, so connect with an admin API key for now.

Google Antigravity keeps its servers in ~/.gemini/config/mcp_config.json, or .agents/mcp_config.json for one workspace. In the editor, choose the menu at the top of the agent panel, then MCP Servers, Manage MCP Servers and View raw config. Antigravity names the address serverUrl, not url.

{
"mcpServers": {
"sealcord": {
"serverUrl": "https://mcp.sealcord.com/mcp"
}
}
}

Antigravity’s docs say it registers itself with a server that allows it, as Sealcord does. Open Agent settings, then Customizations, and choose Authenticate next to sealcord. Sign in in the browser, copy the authorization code it shows, paste it into the settings panel and choose Submit.

Nous Research’s Hermes Agent reads its servers from ~/.hermes/config.yaml, under mcp_servers. Run /reload-mcp after changing it.

mcp_servers:
sealcord:
url: 'https://mcp.sealcord.com/mcp'
auth: oauth

Then run hermes mcp login sealcord from a new terminal: the sign-in opens in your browser.

OpenClaw keeps its servers in ~/.openclaw/openclaw.json, under mcp.servers, or in the Control UI under Settings, then MCP, then Add server. Set transport to streamable-http: without it, OpenClaw uses the older SSE transport.

{
"mcp": {
"servers": {
"sealcord": {
"url": "https://mcp.sealcord.com/mcp",
"transport": "streamable-http",
"auth": "oauth"
}
}
}
}

Then run openclaw mcp login sealcord, or, as an administrator, choose Sign in for the server in Settings, then MCP, and sign in.

A client that implements the MCP authorization specification only needs the URL, https://mcp.sealcord.com/mcp. Its sign-in must redirect to an https address, or only to this computer (localhost or 127.0.0.1).

  • Connect, check or disconnect a payment integration, or see a provider token or a webhook secret. Those stay in the Console.
  • Add, change or remove a webhook endpoint, or replace its secret. An assistant can read endpoints and deliveries and retry a delivery.
  • Create or change products, or rotate a signing key. A wrong rotation breaks apps you already shipped, so use the Admin API or the Console.
  • Anonymise a licence. It cannot be undone, so it needs the licences:anonymise scope, which no assistant is granted.
  • Rename, delete or transfer a workspace. An assistant can list your workspaces, create one you own and move a product between two you own.
  • Change the team without a person. members:write is granted only to an assistant you sign in, never to an API key, and each change runs as you, with your role at that moment.

The server accepts 120 requests a minute per client address. A key the API refused is answered 401 for 5 minutes without being checked again.

Licence keys are secrets that belong to your customers. Only sealcord_get_licence_key and sealcord_issue_licence return one. Anything a tool returns enters the assistant’s context, so prefer the read tools that do not return keys.

The assistant can do exactly what you grant, or what its key’s scopes allow, and nothing more. Grant read scopes for support questions, and ask the assistant to confirm before any tool that revokes, expires, frees a seat or changes your team.

Can an assistant see my customers’ licence keys?

Section titled “Can an assistant see my customers’ licence keys?”

Only if it holds licences:read and calls sealcord_get_licence_key, or issues a licence. The other tools leave keys out.

When you sign in, yes: its actions are recorded as you, through the MCP server, and held to your role. With an API key, they are recorded as that key.

Any MCP client that supports Streamable HTTP servers. Claude Code, Claude Desktop, claude.ai, ChatGPT, Codex CLI, Cursor, Antigravity, Hermes Agent and OpenClaw are covered above. Each connects with your sign-in or an admin API key, with two exceptions: ChatGPT on the web connects only with your sign-in, on the plans OpenAI offers custom MCP servers on, and Cursor only with an admin API key for now, since Sealcord refuses its sign-in. A client that runs only local servers connects through mcp-remote.