Everythings

Connect AI agents

Connect Everythings from Claude's connector directory — or add it as a custom connector in ChatGPT or Gemini — and sign in. The agent can then read and write your workspaces over the Model Context Protocol. In Claude, what it shows you is the Everythings view itself, drawn in the chat.

The one thing you need

Every agent connects to the same address — the Everythings MCP server URL. Copy it, then follow your agent below. The server speaks streamable HTTP and signs you in with OAuth, so in the apps below there's no token to paste. In Claude you can skip even the URL and connect from the directory listing.

https://www.everythings.app/api/mcp

Claude

Everythings is published in Claude's connector directory, so there is nothing to paste — open the listing and click Connect.

  1. 1. Open claude.ai/directory/everythings (or search Everythings under Settings → Connectors in claude.ai or Claude Desktop).
  2. 2. Click Connect.
  3. 3. Click through the Everythings sign-in and approve read-only or read & write access.

On a build without the directory, choose Add custom connector under Settings → Connectors and paste the MCP URL above instead.

Using Claude Code? The Everythings plugin wires these tools into every project and adds /things, a live panel of your workspaces.

Power user: Desktop config file or Claude Code CLI (access token)

Edit claude_desktop_config.json — macOS: ~/Library/Application Support/Claude/, Windows: %APPDATA%\Claude\. Replace nthings_mcp_YOUR_TOKEN with a token from Settings → MCP Access Tokens.

{
  "mcpServers": {
    "everythings": {
      "type": "http",
      "url": "https://www.everythings.app/api/mcp",
      "headers": { "Authorization": "Bearer nthings_mcp_YOUR_TOKEN" }
    }
  }
}

Claude Code (CLI)

claude mcp add --transport http everythings https://www.everythings.app/api/mcp \
  --header "Authorization: Bearer nthings_mcp_YOUR_TOKEN"

ChatGPT

Custom connectors need Developer mode, available on ChatGPT Plus, Pro, Business, Enterprise, and Edu.

  1. 1. Open Settings → Connectors → Advanced and turn on Developer mode (on some builds this lives under Settings → Security & login).
  2. 2. Back in Connectors, click Add custom connector and paste the MCP URL above.
  3. 3. Authorize when ChatGPT redirects to the Everythings sign-in screen.
  4. 4. Turn the connector on inside a chat from the Developer-mode / tools menu.

Everythings also exposes search and fetch tools, so it works in ChatGPT Deep Research as well as chat.

Gemini

  1. 1. At gemini.google.com, open Settings → Connected apps (you may need to pick Personal Intelligence first).
  2. 2. Under Custom apps for Spark, click Add a custom app and paste the MCP URL above.
  3. 3. Click Next and follow the sign-in prompts.

Custom apps for Spark currently require a personal Google account, the US region, age 18+, and "Keep Activity" enabled. English only for now.

Power user: Gemini CLI (access token)

Gemini CLI:

gemini mcp add everythings \
  --url https://www.everythings.app/api/mcp \
  --header "Authorization: Bearer nthings_mcp_YOUR_TOKEN"

Or add to ~/.gemini/settings.json:

{
  "mcpServers": {
    "everythings": {
      "url": "https://www.everythings.app/api/mcp",
      "headers": { "Authorization": "Bearer nthings_mcp_YOUR_TOKEN" }
    }
  }
}

Access tokens

The in-app connectors above sign in with OAuth and need no token. The power-user options (config files, CLIs) use an MCP access token instead. Create one at Settings → MCP Access Tokens, give it a name, and pick a scope.

Read only

The agent can search and read your workspaces, things, comments, and marks. Nothing it does can change your data. Start here.

Read & write

Adds creating, editing, moving, copying, deleting (to Trash), and restoring things, plus comments and marks. An agent can't empty your Trash. Only grant this when the agent needs it.

A token is shown once, right after you create it. Copy it immediately and store it somewhere safe — you can't view it again, only revoke it and make a new one.

What you see

Ask Claude to show a workspace or a thing and the answer arrives as the Everythings view, drawn in the conversation the way the app draws it: emoji tiles, the thing's Markdown, its marks and its comments. No JSON to read.

It's yours to click

Click a tile to open that thing. The breadcrumb across the top — workspace › parent › thing — walks you back up, and the workspace picker switches between workspaces. Every click fetches fresh data through your connection, so nothing on screen is a stale copy. open in app ↗ hands the thing to the full app in a new tab.

You see what an agent wrote

After creating or editing a thing, the agent shows it back to you in this view — the finished thing, in place in your workspace, so you can approve it or ask for a correction while the context is still on screen.

The view never writes

Clicking around calls two read tools and nothing else (get_workspace_view and get_thing_view), so it works on a read-only connection and can't change your data whatever you click.

This is the MCP Apps extension, supported today by Claude. Clients without it — ChatGPT, Gemini, the CLIs — receive the same workspace and thing data as plain text and answer from that, so connecting anywhere still works.

What an agent can do

Read tools (read access)

get_workspace_view, get_thing_view, search_things, get_thing, list_workspaces, get_workspace, list_things, list_child_things, recent_things, list_comments, list_marks, list_trash (plus search / fetch for ChatGPT)

Write tools (read & write access only)

create_workspace, rename_workspace, create_thing, update_thing, move_thing, reorder_things, copy_thing, delete_thing, restore_thing, add_comment, add_mark, remove_mark

Questions

What is MCP?

The Model Context Protocol is an open standard that lets AI agents connect to external tools and data. Everythings runs an MCP server, so any MCP-compatible agent can read — and, with your permission, write — your workspaces and things.

Do I need an access token?

Not for the in-app flows. When you connect Everythings in Claude (from the connector directory listing at claude.ai/directory/everythings), ChatGPT, or Gemini, the agent signs in with OAuth and you approve access on an Everythings screen — nothing to copy. Access tokens are only for the power-user path: a local config file or a CLI (Claude Code, Gemini CLI).

Why does Claude draw my workspace instead of printing text?

Everythings ships an MCP App: the get_workspace_view and get_thing_view tools carry an interactive view that Claude renders inline, so you get emoji tiles, Markdown, marks and comments rather than JSON. It reads your data over the connection you already authorized. Clients that don't support the MCP Apps extension receive the same data as text and answer from that.

Can the rendered view change my data?

No. Clicking a tile, a breadcrumb, or the workspace picker calls get_thing_view or get_workspace_view — both read tools — and nothing else. The view works on a read-only connection, and writes only ever happen when the agent itself calls a write tool.

What's the difference between read-only and read & write access?

Read access lets an agent search and read your workspaces, things, comments, and marks. Read & write additionally lets it create, edit, move, copy, delete, and restore things, and add comments and marks. Deleting a thing over MCP only moves it to Trash (it's reversible with restore), and an agent can never empty your Trash or permanently delete a thing — that stays a human action in the app. The OAuth consent screen and the access-token scopes both let you choose; use read-only unless the agent genuinely needs to change your data.

Is this the same as the AI Provider API key in settings?

No. The AI Provider key (OpenAI/Anthropic) is used by Everythings' own voice and AI features. Connecting an agent over MCP is the opposite direction — it lets an outside agent reach in to your Everythings data. They're unrelated.

Can I revoke access later?

Yes. Connected agents and every access token are listed in Settings → MCP Access Tokens with a revoke button. Revoking takes effect immediately. Treat a token like a password — anyone holding it has that token's level of access.

Which agents are supported?

Any client that speaks MCP over streamable HTTP. This page covers Claude (app, Desktop, Code CLI), ChatGPT (developer mode + Deep Research), and Gemini (Spark + CLI), but the same URL works for any MCP-compatible agent.

    Connect AI agents (MCP) — Everythings | Everythings