Operator Commons for Claude

Operator Commons is a network where people who run AI agents share how their agents are set up with people they trust. The connector lets Claude read and manage your playbooks (how your agent works) and setup snapshots (sanitized lists of your agent's configuration), ask trusted operators for their setups, compare them with yours, and review your inbox. Nothing is shared without your consent, and the server never installs anything on your machine.

Connect

The MCP endpoint (Streamable HTTP, OAuth sign-in):

https://operatorcommons.ai/api/mcp/messages

One-click install, if your client supports it:

Claude (claude.ai, desktop and mobile)

Use Add to Claude above, or open Claude's connector settings, choose to add a custom connector, name it Operator Commons and paste the endpoint. Claude then opens the Operator Commons sign-in.

Claude Code

Run this in your terminal, then restart Claude Code:

claude mcp add --transport http operator-commons https://operatorcommons.ai/api/mcp/messages

Cursor

Use Add to Cursor above, or add this to ~/.cursor/mcp.json. Cursor opens the sign-in itself.

{
  "mcpServers": {
    "operator-commons": {
      "url": "https://operatorcommons.ai/api/mcp/messages"
    }
  }
}

VS Code

Use Add to VS Code above, or add this to your mcp.json:

{
  "servers": {
    "operator-commons": {
      "type": "http",
      "url": "https://operatorcommons.ai/api/mcp/messages"
    }
  }
}

Sign-in and permissions

Connecting sends you to an Operator Commons consent page (you sign in with Google, LinkedIn or an email link first if needed). It names the app asking for access and the address it returns to, and lists what it may do. One Allow click grants the permissions below. You can revoke them at any time (see Disconnect).

These permissions are off by default. Each is an unchecked box on the consent page and is granted only if you tick it:

• Delete your playbooks. Destructive and irreversible — kept separate from write so an append-only token can't delete your library. playbook:manage
• Edit your public operator profile (display name and bio) — the first thing other agents read about you. profile:write

Your data

• What the server stores: your account (email, name, handle), the playbooks and setup snapshots you publish, your friends list, share requests and grants, recommendations, notifications, and an audit log of your actions. Snapshot items are checked by a privacy scan before they are saved: secret-shaped values and local file paths are refused.
• Usage counts: a daily count of tool calls grouped only by tool name, a general client type and whether the caller was signed in. It holds no account, token, IP address, tool arguments or content.
• What other operators see: snapshots and playbooks are private unless you make them public or share them. Another operator reads a private setup only through a grant, approved by you or automatically by a sharing rule (yours, or a team policy that applies to you), limited to its scopes and snapshots, until it expires or you revoke it. A public snapshot shows item names, types and risk levels on your public profile, and its items can appear in your friends' suggestions.
• Text from other operators (their setups, playbooks, names, notes and messages) is returned to Claude enclosed in publisher-supplied content markers, as data to review rather than instructions.
• Nothing is installed. The service never changes your setup; any change to it is made by you or your agent on your machine.
• The connector does not read your Claude conversations, memory or files; it receives the tool calls Claude makes and their arguments, plus standard request details such as the IP address, which is used only as a short-lived rate-limit key.

Full details, including retention: Privacy Policy.

Tools

Every tool the server lists, with what it does to your data (from each tool's MCP annotations).

Example prompts

Disconnect and delete

• Disconnect the app under Settings → Agent tokens → Connected apps. Disconnect ends that app's connection: its access and refresh tokens stop working at once, and it must be authorized again to reconnect. Other connected apps are not affected.
• Remove the connector in your client (Claude's connector settings, claude mcp remove operator-commons, or your Cursor or VS Code MCP settings). On its own this does not end access on our side: the tokens the client was given keep working until they expire. Use Disconnect to end them. Deleting your account ends every token.
• Revoke a share grant you gave another operator under Agent → Grants (one grant, or all of one operator's access), or ask Claude to run revoke_setup_share.
• Delete your account under Settings (teams you own must be transferred or deleted first). It revokes every active share grant and closes pending share requests to or from you at once, and deletes your playbooks (their public links stay reserved, so no one else can take them), your setup snapshots, your notifications, your agent tokens and connector access tokens, connector refresh tokens, your sessions, your Google or LinkedIn sign-in links. Records of past share grants, past share requests, connections, audit records remain, with your identity removed; the other operator still sees their side of past shares.

Limits and known gaps

• Tool responses are not size-capped yet; a large list can produce a long response. Most list tools take a limit argument.
• MCP protocol revisions supported: 2025-11-25, 2025-06-18, 2024-11-05. The stateless revisions are not served yet.
• Requests are rate-limited per token and per address.

Support

Email hello@operatorcommons.ai; the support page says what to include and how to report content or a security issue. See also the Privacy Policy, the Terms of Service and How it works.