In short
- Works with the tools you use.Ready-made setup for Claude Code, Cursor and VS Code, OAuth sign-in for the claude.ai, Claude Desktop and ChatGPT connectors, and any other MCP client.
- Tools scoped by your key.A client only sees the tools its key has permission for.
- Every call is logged.Each tool call is recorded in your audit log.
More and more work happens inside AI assistants. MCP, the Model Context Protocol, is the open standard those assistants use to connect to other services. OneType is an MCP server, so you can ask Claude Code or Cursor things like which conversations went unresolved yesterday, or what your agent says about refunds, and get the answer without opening the dashboard.
What your AI tool can do
- list_agents and get_agent: see your agents and their settings.
- list_conversations, get_conversation and list_messages: read conversations, filtered by agent or status.
- search_knowledge: find the best-matching passages from an agent's pages, files and Q&A, with their source.
- ask_agent: ask an agent a question as a visitor would, and get its cited answer.
The answers are the same objects the REST API returns. Only ask_agent is billed, like a chat reply: AI credits on Premium. Premium with your own AI key, which will count answers instead, is coming soon.
Connect in three steps
First, create a restricted key in Developer, API keys, and tick only the permissions for the tools you want. Second, open Developer, MCP, and copy the snippet for your client: there are tabs for Claude Code, Cursor, VS Code and curl, with your address already filled in. Third, ask your question in the editor.
A tool only appears when the key holds its permission. A key that can only read agents will never see conversations. Give each client its own key, so you can revoke one without touching the others.
Safe for AI to read
Conversations and knowledge are written by visitors and by your own content, so every passage, message and answer is marked as untrusted text for the calling model. That tells the assistant not to follow instructions hidden inside them. A tool result never starts anything on the server, every call is recorded in your audit log, and requests sent from a web page are refused, because keys belong on servers and desktops.
Details for MCP client builders
The server speaks Streamable HTTP at /v1/mcp and supports protocol versions 2025-11-25 and 2025-06-18. It answers ping, tools/list and tools/call. Sessions end after 24 hours without a call, and a client simply starts a new one.
Useful questions to try
- Which conversations for my support agent are unresolved?
- Show me the messages in the last escalated conversation.
- What passages does my agent find for how do refunds work?
- Ask my agent how to change a billing address, and show the sources.
Your assistant picks the right tool, calls it with your key and answers in plain words. Because search results come with scores and sources, it is also a quick way to check whether your agent finds the right content.
Anyone with the key can use it, so treat it like a password and keep it out of shared config files.
What each plan includes, and how AI credits work, is on the pricing page.