Appearance
Cursor, VS Code and other MCP clients
Clients that are configured with a file or a command, rather than an "add a connector" screen, connect to Subtext with an access key. The client sends the key as an HTTP header on every request.
- Endpoint:
https://mcp.subtext.wiki/mcp - Transport: streamable HTTP
- Header:
Authorization: Bearer <key>
If your client supports OAuth 2.1 for remote MCP servers, you can skip the key and point it at the endpoint alone; it will discover sign-in from https://mcp.subtext.wiki/.well-known/oauth-protected-resource/mcp and ask you to sign in with Google.
Create an access key
- Open the Connections page in the Subtext app.
- Choose Add a key.
- Give it a name for the tool or place you'll use it, such as
Cursororwork laptop. The name is how you'll recognise it later. - Choose How much can it see?
- Just me: only your name and role.
- Working context: you, your organisation and your active projects. The usual choice for a day-to-day work assistant.
- Full context: everything in your graph, including people, tools and topics.
- Choose Can it make changes?: Read only or Read + write.
- Choose Add key, then pick your client from the tabs to see the exact setup for it.
The key is shown once. Copy it into your client straight away; Subtext stores only a hash of it and cannot show it again. If you lose it, make a new key and revoke the old one.
About keys
- One key per connection. Make a separate key for each tool or machine, so you can see which one is reading and cut one off without affecting the others.
- Each key has its own scope and write setting. You can change both later from the key's page, and the change applies to the next request.
- Each key keeps a record of its reads. The key's page lists what it searched for and which pages it read, with times.
- Pause and revoke. Pause stops the key from reading anything until you Resume it; the assistant is told briefing is paused. Revoke key deletes it permanently, and any client still using it is refused from then on.
- Writing needs the right key. A Read only key can't change anything. Submitting new information with
edit_page_bodyalso needs Full context, because the update can touch any page it mentions.rename_pageandlink_pageswork with a narrower Read + write key, as long as the pages are within its scope.
Treat a key like a password: keep it out of shared repositories and screenshots.
Cursor
On the Connections page, the Cursor tab has an Add to Cursor button that opens Cursor with the server filled in. To add it by hand, put this in .cursor/mcp.json in your project (or ~/.cursor/mcp.json for every project):
json
{
"mcpServers": {
"Subtext": {
"url": "https://mcp.subtext.wiki/mcp",
"headers": {
"Authorization": "Bearer <key>"
}
}
}
}VS Code
On the Connections page, the VS Code tab has an Add to VS Code button that installs the server for you. To add it by hand, put this in .vscode/mcp.json in your workspace:
json
{
"servers": {
"Subtext": {
"type": "http",
"url": "https://mcp.subtext.wiki/mcp",
"headers": {
"Authorization": "Bearer <key>"
}
}
}
}Note that VS Code uses servers and needs "type": "http", where Cursor uses mcpServers.
Claude Code
bash
claude mcp add --transport http "Subtext" https://mcp.subtext.wiki/mcp --header "Authorization: Bearer <key>"See Connect Claude for the OAuth alternative and for installing the skill.
Any other MCP client
Most clients that support remote MCP servers need the same three things:
| Setting | Value |
|---|---|
| URL | https://mcp.subtext.wiki/mcp |
| Transport | Streamable HTTP (sometimes labelled "HTTP") |
| Header | Authorization: Bearer <key> |
If a client asks for an address and a key in separate fields, paste the endpoint as the address and the key on its own (without Bearer) as the key.
Keys created on the Connections page start with cek_.
The skill in other clients
Clients other than Claude generally don't use skills; they rely on the server instructions and tool descriptions Subtext sends when they connect. See The Subtext skill for what those cover.