pergrn hosts a remote MCP server at https://pergrn.com/api/mcp (Streamable HTTP, OAuth 2.1). There is nothing to install on pergrn's side and no API key to copy — clients discover the login flow automatically.

claude mcp add --transport http pergrn https://pergrn.com/api/mcp

On first use, your browser opens pergrn: sign in if you aren't already, review the consent screen, and choose Allow access. From then on you can ask things like “list my pergrn projects” or “add a note to Garden Planner”.

Any client that supports remote MCP servers over Streamable HTTP can connect with just the URL:

{
  "pergrn": {
    "url": "https://pergrn.com/api/mcp"
  }
}

Clients that only speak stdio can bridge with mcp-remote:

{
  "pergrn": {
    "command": "npx",
    "args": ["-y", "mcp-remote", "https://pergrn.com/api/mcp"]
  }
}

• list_projects — the projects you own or joined.
• get_project — one project's metadata: description, status, and its repository / live / docs URLs.
• list_notes / get_note — read a project's notes in your arranged order.
• create_note — add a note to the top of a project.
• update_note / complete_note — edit content or mark notes done.
• delete_note — permanent, and clients are asked to confirm before calling it.
• set_project_repo — set or clear a project's repository URL.

Changes made by agents appear live in your dashboard and for your collaborators, exactly like edits you make in the browser.

• A connected agent acts as you: it can read and edit notes in every project you can. Treat connecting an agent like handing over your own keys.
• Consent is per client and recorded against your account. Nothing connects without the consent screen.
• To disconnect, remove the server from your client's MCP configuration; the client's access stops working when its tokens expire.
• Every request is authorized per user by row-level security in the database — the same enforcement your browser session uses.
• You can review and revoke connected agents any time in Settings → Connected agents.

Agents without a browser — cron jobs, CI, agent loops — authenticate with a personal access token instead of the sign-in flow. Create one in Settings → Agent access; it is shown once, and you can revoke it there at any time.

{
  "pergrn": {
    "url": "https://pergrn.com/api/mcp",
    "headers": { "Authorization": "Bearer pgn_…" }
  }
}

A token acts as you, under the same row-level security as everything else. Prefer an expiry, and treat the token like a password.