Public machine-readable access

Trust Onion for AI agents

Give compatible Claude, ChatGPT, Cursor, Gemini CLI, and other MCP clients a live connection to verified AI scam incidents, current scam guidance, and the Three Words Check protocol.

Install the MCP server OpenAPI specification

Remote server URL

One endpoint

Use this URL wherever your agent host asks for a remote MCP server:

https://trustonion.io/mcp

A direct browser or crawler fetch returns machine-readable connection details. MCP clients should advertise text/event-stream and complete the JSON-RPC initialize handshake; this is not a normal JSON API endpoint.

No account, API key, or write permission is required. Public requests are rate-limited.

Claude.ai / Cowork

Use the Connectors Directory

In Claude, open the + menu in a chat and choose Connectors, then select a connector from Anthropic's directory. Claude does not expose an arbitrary remote-MCP URL field in every account.

If your account does not show Add custom connector in its Connectors settings, you cannot add Trust Onion by pasting its URL. Team and Enterprise workspaces may require an owner or admin to configure connectors. See Claude's connector documentation.

ChatGPT

Create a Developer Mode app

ChatGPT's normal app directory is not a general MCP URL importer. On eligible web accounts, turn on Developer mode in Settings, then use the Apps or Plugins + control to create a developer-mode app from this remote MCP URL. Enable the app in the chat composer before using it.

Business, Enterprise, and Education workspaces may require an administrator to enable Developer mode. If Developer mode is not available, this account cannot add an arbitrary remote MCP server. See OpenAI's current setup guide.

https://trustonion.io/mcp

Cursor

Paste into MCP settings

Open Cursor's MCP settings and add a remote server. Cursor supports remote HTTP and SSE servers; save the configuration, restart Cursor if prompted, and enable Trust Onion in the MCP panel.

{ "mcpServers": { "trust-onion": { "url": "https://trustonion.io/mcp" } } }

The same entry can be added to Cursor's project or user MCP configuration. See Cursor's MCP documentation.

Gemini CLI

Add it to settings.json

This applies to Gemini CLI, not automatically to the Gemini web app. Open ~/.gemini/settings.json or the project’s .gemini/settings.json, add the remote HTTP server below, and restart Gemini CLI.

{ "mcpServers": { "trust-onion": { "httpUrl": "https://trustonion.io/mcp" } } }

Gemini CLI also supports an url field for SSE servers. See Gemini CLI's MCP documentation.

Tools and resources

Discover the live catalog

After the initialize handshake, use the MCP tools/list and resources/list methods to discover the current read-only catalog. A direct fetch of /mcp returns the same connection metadata as machine-readable JSON, including the available tool and resource names.

The catalog is generated from the live MCP server metadata, so agents should not hardcode a separate list of tools. Resource contents are read from the current published database and topic guidance.

Live resources

Current content, not a deployment snapshot

Each request reads the current published database and topic guidance. Agents can also read these MCP resources:

Attribution and freshness

Cite the source

Every tool result includes an asOf timestamp, a content revision timestamp when source rows are dated, and a canonical Trust Onion source URL. Incident results also include original reporting links when available.

This endpoint is read-only. It does not send messages, create accounts, publish content, or make an external AI call to assess a scenario.